Dreaming about Better Sleep: Dreem, Oura and the Rest

Tomáš Baránek
Jul 14, 2019 · 51 min read

Perhaps it’s already clear to my readers from the number of articles I have written on sleep that I’ve been trying to find my way through this “hall of mirrors” of my life for a long time. Since the times of diapers; or maybe it would be better to say since the times of sleeping in sleeping bags. I will never forget the constant waking up, the long hesitation, the childhood fear, the chill in the air and the silent envy that the others were still asleep — in summer camps and on trips, when I had to leave the warmth of my sleeping bag several times a night to go out and pee somewhere in the dark. It’s still a familiar feeling up to today, only I’m just worrying about what the next day will be like, not about wild boars.:) Sleep is extremely important, it’s the pillar on which our physical and mental performance depends, our health, regeneration, metabolism, the speed at which we age…

And yet, as Professor Matthew Walker says, 80% of people in the West are suffering from sleep deprivation today. According to the number of grim stories of tired friends and loved ones who boast on the one hand that they “don’t need to sleep more than 6 hours”, and at the same are putting on weight or have various chronic health problems, I would say that his estimate is still conservative. Further proof of this can be seen not only in the extraordinary success and response to Walker’s bestselling book Why We Sleep (+ the buzz around it), but also by the growing number of gadgets and apps that have been developed to measure, evaluate and even influence sleep.

In this fairly lengthy text, I want to present the results of a month-by-month comparison of several state-of-the-art sleep screening devices (I refer to them simply as sleep trackers) that are intended to be used at home.

REM, NREM and Everything Around Them

First of all, let’s recapitulate the terminology and some basic facts. Healthy sleep has a typical multi-stage architecture. These are NREM1 to NREM4 (NREM1 and NREM2 are light sleep, NREM3 and NREM 4 range from deep sleep to very deep sleep) as well as the REM phase. These cycles take turns after about 90 minutes and cleverly follow each other. The rest of the night is filled with wakefulness. The time from switching off the lights to falling asleep is referred to as latency.

Sleep is initiated by at least two major processes: fatigue (sleep pressure, the adenosine cycle) and the excretion of melatonin into the blood (the Circadian rhythm). These two cycles should be in sync with each other. If they diverge regularly, for example, if you don’t have enough melatonin in your blood before falling asleep, problems will start to accumulate. Initially, they will be creeping and manageable and probably not attributable to insufficient sleep, but these will grow into chronic health problems over years and decades.

But for sleep to be sleep, and not to just be some state of unconsciousness, and for it to perform a variety of healing and key functions for the body and brain, it must be preceded by a prelude throughout the entire day and must have the right timing and length. When it comes to assessing the quality of it, it is not just about length, but about the depth and quality of each phase.

Plenty of people say that they “have no problem falling asleep in front of the television,” unfortunately this is a bad sign; they are falling asleep in a bad place. If you spend your days in an office under fluorescent lights and your evening glued to a screen until the final moment, it is true that you fall asleep because of being tired, however you won’t be able to rest at night (due to the lack of melatonin through which our internal organs typically recognize that it is night and time to rest). Along the same lines, the idea of “healthy drunk sleep” is a myth, because drinking alcohol is not about getting high-quality sleep and the brain and body are not able to regenerate, just getting by in a kind of unconscious state. Of course, this isn’t only about alcohol — caffeine, various foods, our level of hydration, the (ir)regularity of lying down and getting up, exposure to and lack of exposure to different types and intensities of light throughout the day, physical activity, health problems, medicine, smoking, etc. can also have a similar negative effect on our ability to sleep. All these circumstances affect the phases, depth and overall architecture of our sleep.

We, the sleep tinkerers, have already known this for quite some time. We try, often in quite an amateur way, to get the particular phases and the quality of sleep during them a little bit more under control or to at least measure them and look for “correlations”, or possible connections, to track the relationship between our lifestyle on the one hand and the quality of sleep on the other. I’ve been trying to do this for years myself, since I discovered the first gadget that promised to measure sleep in 2010 — called Wakemate — which completely fascinated me. Unfortunately, it didn’t work very well, and this endeavor later went on to languish, like so many other similar projects that promised to work like magic.

How Sleep Is Measured in a Sleep Lab

A professional sleep laboratory uses a so-called polysomnographer to scan several (12 or more) bits of biometric data from someone while they are sleeping, the main ones being: EEG (brain waves), ECG (heart activity), EOG (measurement of eye movement), and EMG (muscle activation), as well as the level of blood oxygen saturation and more. The different secondary quantities are then derived from these directly measured values ​​and the result is a polysomnogram which shows dozens of values ​​that can be used by the sleep expert for comparison and evaluation.

A polysomnogram — above you can see the brain waves, then below the other metrics. Source: Wikipedia.

On the basis of these metrics, the individual pieces of sleep are then evaluated and, among other things, a hypnogram is created, i.e. a graph of the night’s progress with estimates about each phase and other sleep parameters. However, the key variables monitored by the experts include brain waves measured using electrodes on the head (EEG) — without them, they cannot and aren’t intending to conduct a professional evaluation.

A hypnogram is the result of the polysomnogram being interpreted by an expert (the polysomnogram shown above does not correspond to this hypnogram). Source: Why We Sleep by Matthew Walker, Simon & Schuster, 2017.

You may suspect that ordinary sleep trackers (a smartphone under the pillow, bracelets, watches, a mat under the bedsheets, etc.) that are not able to measure any of the primary variables, rely on secondary or derived phenomena (sleep movements/actigraphy, heart rate, HRV) and determine sleep phases through a so-called heuristic technique. These are guesstimates. In other words, they speak inaccurately and often inconsistently about the stages of sleep, because the heuristic here is based on only a few pieces of (secondary) biometric data and therefore is more prone to bias by circumstances (e.g. the age of the sleeper, their long-term, as well as acute, health conditions, etc.). It should be said that the manufacturers of many gadgets and applications indirectly acknowledge this deficiency by calling the phases of sleep only “light sleep,” “deep sleep,” and “awake”. If anyone says their gadget or even app identifies your REM, NREM (or its sub-phases), be on guard. Without an EEG it will not work, as I will show in the comparison at the end of the article.

If anyone says their gadget or even app identifies your REM, NREM, be on guard.

Nevertheless, I think that even relatively inaccurate data can be used as a sort of orientation, as a rough guide and gateway to the world of self-measurement and sleep-hacking. The majority of apps for basic sleep tracking moreover will also force you manually adjust the time when you fall sleep and wake up, so that if nothing else, you can completely keep track with the app of the exact time you fall asleep and wake up, that is, the so-called sleep opportunity. Thanks to these records, complete with a simple rating for the night and the next day, you can develop a better awareness and be more motivated to work on your sleep habits. Gradually, your interest in having more accurate data and information will grow, or perhaps you will quickly come to the conclusion that it is better to go to a doctor. If, despite all your efforts and without a known cause, you suffer from severe insomnia or excessive fatigue during the day, a sleep laboratory could be the answer. I have been on the verge of going there lately…

A Short Anecdote — or How I Doubled My Amount of Deep Sleep

The following story is important because through it I want to show you how difficult it is — even with the greatest amount of effort, looking into quality resources and racking up considerable expenses on measurements — to find the origin of one’s sleep problems. From last December to March of this year, my sleep was terrible. Every morning, without exception, I felt drowsy and had brain-fog, and without a good cup of coffee and tea I couldn’t function. I planned my days so that I could take a short nap in the afternoon and refresh myself at home or at work (where we have the opportunity to do so). A midday nap is probably good for the heart, but it shouldn’t be something that you feel the need for every day or that you wouldn’t survive the afternoon without it.

I had already gotten fed up with these complaints and the constant analysis of the causes was annoying to me and the people around me, but — as I can see today — the overall quality of my life was really falling. At night, everything was wrong, according to Oura (I have my reservations about the accuracy of its measurements, but I will describe this further in the review below), and also according to my subjective “memories of the night.” I was waking up more and more at night, went to the toilet 3 to 4 times per night, couldn’t get up in the morning, etc.

Although this “insomnia campaign” has happened to me in the spring for a number of years, it had never lasted for weeks and nor had such intensity. It was only one night in April after which I forgot to make coffee and my head was running at 100% performance all day long, that I began to have hope that normal life still existed somewhere. I tried to remind myself of everything I had done differently in the days and evenings better than before.

My deep sleep (purple graph) and sleep efficiency as a ratio of quality sleep for lying in bed (red graph) since October 2018 to the present. This was measured by Oura, my reservations are reviewed below.

What was the turning point? Afterwards a few worse nights followed, but gradually the situation (unbeknownst to me) began to improve significantly until I began to sleep acceptably. In the morning I stopped feeling the heavy sense of numbness and drinking coffee began to be voluntary. And of course: I saved some time by no longer needing a midday nap. (These days I can understand again the questions other people kept previously asking me: how can you fall asleep at noon?). However, the improvement was not only subjective. Oura claimed that I had a lower degree of restlessness, light sleep and my average deep sleep length improved significantly (2x, 1.5 hours). Incidentally, my Heart Rate Variability (HRV) rose and my average heart rate fell, but I’ll get to this later.

Since then, I have been praying daily that the trend will not turn around and my wife laughs at me now that I must continue to do all the measures that I have taken to gradually improve my sleep. The cause of the improvement is probably not limited to just one thing, since it’s probably more than one factor.

What beneficial sleep measures have I gradually introduced? I’m afraid that there are more than a few!

For More than Six Months:

  • After 9 pm I wear red glasses, and from 6 pm in the evening we dim the lights to a warmer setting or through the second degree of the DEN light switches.
  • Where I sleep, I always maintain silence (with earplugs) and darkness (either through the use of blackout curtains, an eye mask, or both) at night.
  • I take magnesium malate or triglycinate 3 times a day.
  • In the morning I took a daily supplement of vitamin D in drops (more details below).
  • I have been taking probiotics for a long time (Lactocare in the morning, as well as Probiolact at noon and in the evening).
  • I drink my last caffeinated drink at 3pm, and have a maximum of 3 per day.
  • I try to go to bed at the same time with religiously (I turn off the lights at 11:15 to 11:30 pm).
  • I always wake up at the same hour (my alarm is set for 8:00) and although I often wake up much earlier, I simply “make myself sleep”, which oddly works — every bit of REM sleep is good.
  • Before 8 am I am not exposed to any blue light.

Two Months Before My Improvement:

  • I drink my last caffeinated drink at 2pm, and only have a maximum of 3 per day.
  • I have reduced my average daily dose of Zoloft (an antidepressant and anxiolytic medication that manipulates serotonin and dopamine, but needless to say it most likely has some effect on one’s sleep); I take it in the morning to minimize the possible immediate impact; however, the changes of these doses manifest themselves slowly, according to recent research, for six months or even up to a year.

2 to 4 Weeks Before My Improvement:

  • I started to expose myself right after waking up (8: 00–8: 10) to blue light (either sunlight, DEN light, or Philips Energy Light…).
  • After a discussion with Jirka Hubík (who said something more or less like: “When I couldn’t sleep, I started to actively do sports every day and now I sleep well.”) I actively incorporated a much larger range of aerobic activity into my program (beside going to the gym once a week and doing yoga once a day) I added skipping rope daily until I was exhausted, going for a run once a week, speed walking several kilometers several times a week, etc.).

A Few Days Before My Improvement:

  • I stopped taking vitamin D (because it occurred to me that it might be a good idea to have it tested and it turned out that the amount of vitamin D in my blood was closer to the upper limit than the lower limit when it was tested).
  • I stopped eating certain foods in the evening: tomatoes, cheese, cabbage, salty stuff and some other foods (that disrupt sleep) and after 7 pm I avoid, for the most part, carbohydrates and food entirely, so that my body and brain can actually slow down; I prefer to go to sleep feeling a little hungry rather than saturated or stuffed.
  • I started to drink a little more during the day, but I stopped drinking fluids around seven, with one exception (see below).
  • Just before going to bed, I drink about 50 ml of solution from two bags of Blockurima (D-mannose), which should somewhat relieve the urinary tract (repeated tests have ruled out any infection).

And so somehow after all this I started to sleep much better. But you might ask: what was really the cause for the improvement and what did not contribute to it? What could I leave out of these crazy routines, what could I do differently? Couldn’t it just be that amount of natural sunlight has increased as the days lengthened? Is it just a matter of discipline and regularity? Or was it the uncompromising few months of omitting any blue light before going to bed? Or perhaps I’m just running around outside more due to the weather getting better? Perhaps my urinary tract has finally gotten better? Has my level of stress been reduced after the winter?

I still wanted to know more. This brought me back to the core of the matter: on one hand, this comes down to how one feels after the night, on the other hand, there are objective pieces of data that often “do not fit” with our immediate intuitive estimates and may have a negative impact later. To get closer to the most accurate knowledge, I got the best thing you can get when it comes to sleep. Dreem 2.

Dreem 2 — So You Know When and How You Really Sleep

I Don’t Look Like This Even with Dreem

This device belongs to the top of home sleeper trackers in measuring sleep accuracy, and I should start off by saying that I use it as a standard by which I can compare the results of other such devices. Dreem 2 started to be sold earlier this year, collecting prizes and positive reviews along the way. I will go into greater detail about my many months of experience later, but for now let me share my suggestive impressions from my May testing, after about 5 weeks (of which the first four are included).

The main parameter in which Dreem significantly exceeds the rest of the gauges is the fact that its sleep phase estimates are based on the measurement of brain activity using six EEG sensors supplemented with a heart rate sensor and a motion sensor (out of which it is then able to derive the respiratory rate, body position and other variables). It uses deep learning algorithms [1] that run directly during one’s sleep in real time in the headband to evaluate it. Why in real time? Because the second, but not least, distinctive factor is the possibility of stimulation during the NREM phase — in the form of sound waves to increase the amplitude of spindles (doses of particular brain waves that last less than a second), the importance of which is crucial for quality sleep.

It is a headband that you wear for the night and take off in the morning and put on its charger. The very concept of a headband with EEG capabilities is far from new and it must be said that other similar devices (e.g. Zeo) have not been successful. Probably because their users were ashamed to wear it on their heads in front of their partners. Most likely because of this, the much more comfortable and more sexy devices — regardless of their poor accuracy — have taken root. The French project Dreem is an attempt to overcome this problem: the headband design is more futuristic than “medical” or something for the elderly. But this isn’t the main thing here.

In addition to the Dreem’s unprecedented accuracy and stimulation effect, what is important about Dreem is the clarity of the application and the breadth of personalized services and advice it makes available to users who are struggling with sleep issues. Dreem works as a long-term sleep expert and one-on-one coach: it monitors your sleep habits and sleep quality for a week, asks you for a lot of more details about your life (day and night), and downloads your activity data from other personal health apps. After a week it will analyze all of this data and generate a detailed sleep report for you. It evaluates your “sleep type”, suggests a course of action for sleep and other measures, among which the headband itself plays an important role. For those who are interested, you can see my first Sleep Report here — I doesn’t make any sense to insert it directly into this article, since it’s basically a long string of data composed of sequential screenshots from my iPhone.

The recommended measures are a combination of acquiring knowledge on sleep habits, well-proven cognitive-behavioral sleep methods, other recommendations for coping with insomnia, guided mindfulness lessons during the day as well as before sleep, napping and also activation of the mentioned stimulation during deep sleep.

The Role of Stimulation for Better Sleep

So, I would guess that many of you are primarily interested in the stimulation part of this. Me too. I have studied the complete white paper almost word for word to which I have been referred by its creators, and as a computer science graduate and a self-styled geek, I have to take my hat off to the developers of Dreem. They have faced countless challenges during which they obviously did not give up and exerted more and more effort in collaboration with sleep scientists. I was intrigued by the high level of accuracy of the sleep phase diagnosis, which is comparable to a classic polysomnograph (90% specificity for NREM), promising efficacy of the stimulation in deepening the spindles, but also by the fact that Dreem is gaining a decent volume of (anonymized) data from Dreem’s users about the immediate effects of the stimulation. Data from a professional laboratory, where such stimulation is examined is perhaps technically more accurate, but not by much and, moreover, it is biased by the problem of the influence that the experimental environment can have on the sleepers.

How exactly does this stimulation work? During the appropriate NREM subphase (mostly during NREM3) the headband generates precisely timed doses of “white noise” that are transmitted to the auditory nerves by vibration from a special vibrating device placed on the outside of the headband. This is not an electrical or other type of signal, it is of a mechanical nature — if you were awake, you would hear a chirp or a whistle (see below). The timing of each dose is therefore absolutely critical: it must take place when you are in deep sleep (otherwise it would wake you up) and the pacing of it must occur at tenths of a second, exactly when a short sleep spindle starts to rise: only then can the stimulation have an effect. The fact that a computer that evaluates all this in real time, is as small as a box of matches, and is completely offline, is totally astounding to me.

After the first night of stimulation, my wife Katerina pointed out to me that soon after falling asleep, my head whistled quietly. (She says it doesn’t disturb her and that it calms her down, but it is necessary to reckon with the fact that a partner near you may hear something.) If it weren’t for her confirmation, I wouldn’t believe the thing was actually making any sounds. Although I have been receiving about 300–400 “doses” every night for about a month (the creators of Dreem call them “stims”), I don’t remember ever hearing a stimulation sound. While this is only anecdotal evidence, it is still solid proof that these stims are really being enacted when I am in deep sleep.

What are the real effects of this stimulation? Subjectively, I feel like I get better sleep in a shorter amount of time, but that may well be the long-term trend of the improvement I have described above, and which has already started before the introduction of the stimulation. My deep sleep has more or less been kept on the same level. What is more interesting is that the REM phase has been gradually increasing since the start of stimulation. The REM phase with deep sleep, therefore, has increased at the expense of light sleep and the times when I’m literally wide-awake.

Let me touch briefly on the technology of transferring the sound through bones all the way to the inner ear. This has its well-known pros and cons. The main advantage is that you do not need to have anything in your ears, the disadvantage is the relatively weak sound (or if need be a better sound while sensing a kind of tingle). For Dreem, this is not only used for sound stimulation, but primarily for verbal coaching, meditation and informing the user. If you press a button on the headband at night, you will hear a woman’s voice telling you that the measurement is still in progress and it will be ready when the time comes (according to the set alarm). Upon waking up (which I am not experiencing nearly as much, since I cannot get back to sleep in the morning), you can use earplugs, yet Dreem will wake you up with a melody or a particular sound. You have the impression that the voice or sounds come as if from inside your head, which is kind of cool. :)

The first nights with Dreem were a bit awkward: I kept worrying that it would fall off my head, it took me awhile to get used to the procedure for turning it on. In the morning I had bruises on my temples and even woke up with a slight “hair” pain, like the one you get from wearing a tight-fitting baseball cap. After about four days, the trouble disappeared, the headband had adapted to my head shape (and I to it), so I basically hardly even realized at night that I was wearing it.

The Dreem App Is also Unique in Many Ways

Dreem’s app is a bit unusually organized — the data is displayed using cards that stack up on your “sleep lines” so that the latest news and information are at the top. The morning briefing begins by transferring data from the headband and drawing up a report from the night:

If you want to take a closer look at how the night went minute-by-minute, tap the report tab to get to the detailed graph. It’s a nice detail to know exactly in what position you were sleeping in particular moments during the night.

On the other hand, if you want to see the information from a greater distance, you can dig through the markers, where it is possible to go through the last week’s results. A weekly view to someone (such as me) may not be enough, but I assume that Dreem will continue to work on developing the app:


  • The accuracy of measurements is amazing: REM, NREM, light sleep and wakefulness are identified by Dreem more accurately than anything I have ever encountered. You could argue that I would have to compare it with a polysomnogram and you would pretty much be right: it is still an estimate, albeit a quite qualified one. I can say that by providing me with hundreds of nights and morning records and analyzes with different trackers, Dreem basically has given me the missing pieces of the puzzle that I had been missing in the evaluation for my first good (night of) sleep; the results before Dreem had always contained strange contradictions in relation to the subjective experience, but I rather attributed this to that particular experience; now the measurements and the realities of the night fit together both in the short and long term.
  • Dreem has the ability to recognize the exact time when you fall asleep and wake up; all the other devices I have ever used regularly have had a problem with knowing that I am asleep, or that I am not sleeping, even though I wasn’t moving; or if you read an hour before bedtime and in the morning you learn that you had been sleeping for that entire hour; you get up 4 times out of bed at night only to be praised in the morning that did not wake up one single time; or that after you were woken up in the morning because of some racket and you have been trying hard for several or dozens of minutes to fall asleep, the app congratulates you for having been in REM sleep for the last two hours of sleep (some apps don’t even allow you to change this data…) — this is the reality of all of the most common methods (apps, bracelets, rings, blankets, etc.) that I have encountered and that I review later in this article; But what of it? If a device can’t identify when you fell asleep and when you wake up, how can it even begin to say that it will be able to recognize your sleep phases? Well, it just can’t…
  • Its Sleep Assessment and Dreem Coaching have a well-built and individualized dramaturgy that is written in witty and perfectly understandable English — a good adviser and companion for both the layman and the intermediate user. For many sleepless people who are not well versed in sleep studies, the crucial step towards their first improvements is simply the introductory “Basics Easy” course — yet there are a lot more and deeper levels that you can explore.
  • The application is clear and easy to navigate — considering how complex service it provides to the user you can understand quite quickly how to work with the app. In fact, it’s clear that the creators had to eat their words and leave out a lot of additional information that they probably had felt the urge to communicate to the user — but they kept themselves in check and kept the message simple — only in that way would a normal insomniac want to use Dreem for a long time.
  • What is unique is the graphic comparison of your sleep with the average of other sleepers using Dreem so you can see how you’re doing. In fact, I was not very comforted by this: in terms of deep sleep I am above average, but otherwise I am stuck around the middle to below average, even though my radar graph has been gradually getting bigger. It’s a form of motivation to work on yourself.
  • Stimulation probably works, and in time it may prove to be a big thing — I personally feel the results regarding hard data already after a few weeks, I have been able to reduce the time I need to sleep in order to have a great energetic day; but in order to not make downright bad scientific observations, I have to watch everything for a few more months, play with other variables and then temporarily turn off the pacing and then turn it on again. Of course, even then it won’t be a test that can guarantee that there will be 100% positive effects for someone else.
  • I find it very handy to listen to the sleep soundtrack when I am falling asleep with an automatic switch off when one falls asleep. The thing is, if you let some murmur of the sea or the sound of rain play on your ordinary headphones before bedtime, the headphones will put pressure on your ears, or they will eventually fall off, or the sound will turn off too soon, or it will keep playing unnecessarily after you have fallen asleep and it might wake you up later on. With Dreem this is not a concern: if you turn on this feature, the selected noise will gradually fade to silence as you fall asleep (not just when you stop moving or your heart rate drops). Ingenious.


  • For some people, it may be limiting that everything (the application, navigation and coaching) is only in English or French; for someone like my mom, who only understands Czech and Slovak and could really use something like this, this means it is pretty much unusable. I would have to create some type of detailed manual (which couldn’t be longer than this article…, so I will consider doing it :).
  • For someone it may be inconvenient or awkward to sleep with a headband; it has only fallen off of my head once, which is quite a good result (sadly that night was simply not recorded at all); Dreem isn’t really all that tight, but the truth is, it still takes a few days to get used to it.
  • The bone conduction audio is of relatively low intensity; it is fine in a quiet bedroom, but if, for example, you want to meditate in a car’s passenger’s seat or in another noisy environment, you won’t be able to do it without the earbuds and even like that you will have difficulties hearing the voice guidance; there is a headphone jack on the headband, but I haven’t had headphones with a classic connector for a few years…
  • The set of background soundtracks for falling asleep or meditation could be more varied and more configurable, but I am treating this as a basis for further development.
  • As I indicated above, I miss there being a more sophisticated way of tracking any longer-term trends and tagging days/nights to make more comfortable comparisons as it is the case of some other gadgets (especially the Oura or Go2Sleep). In the application it is possible to see the measured values only separately, one below another, one graph after another graph, week after week. Fortunately, it is possible to export this data, but not with too much detail; a web interface like the one that Oura uses, would move the whole thing for the more experienced self-trackers to the next level, higher (and especially deeper).
  • Since the headband has EEG data, it would seem to me that the device might be able to detect in time sleep apnea, which is a common and dangerous (and often hidden) sleep disorder; perhaps one day Dreem will be able to recognize this problem through deep learning.
Light spectrum analysis of Dreem LED.
  • I was disappointed by the lack of attention with which led the creators to ignore the negative effect of blue light on sleep (although they point out this issue in the coaching). The application itself radiated a white and blue light during the night, as if this doesn’t matter (and they don’t even warn the user). What’s more: The LED on the headband flashes white and glows blue. Meaning that when you go to sleep, a device that is intended to deepen your sleep exposes you to blue light with a frequency culminating at 460 nm (measured with a $3,690 spectrometer, the UPRtek 350 S Premium). This is truly a shame. The trend of using blue LED lights must end and I have already written to the creators. They didn’t exclude the possibility of an improvement via a remote firmware update.

The Verdict: Dreem is an excellent and very promising device out of the latest generation of such devices, which leaves the others behind in terms of accuracy of measurements. Of course, I cannot prove that the perceived improvement in my sleep that has been slowly taking place is the result of using Dreem — it just seems to be the case so far. If you are in a situation where you have been struggling with sleep issues for a long time and have 399 EUR that you can spend on your health, then it’s worth trying Dreem since it has a real chance of improving your sleep habits and your ability to sleep without taking any medications.

The Elegant, but Inaccurate, Oura (or the One-Eyed Man Among the Blind)

FYI this isn’t me wearing an Oura ring too

Before I started to use Dreem, the main device for measuring my sleep was an Oura Ring (2nd generation, for brevity’s sake I’ll refer to it from hereon as simply: Oura). From the beginning, Oura has provoked enthusiasm in me, and in some respects, this has continued right up to today. When you see it for the first time, it’s hard to believe that this little shiny bit of nothing could “do something”: it’s just so small. Oura is a ring, so (after choosing the right size during the purchase) you wear it on one of your fingers — it doesn’t have to be worn during the day, but understandably you have to at night. Although I don’t wear rings, I got used to it: when people gave me a weird look, I would show them the chip from the inside; it’s pretty easy to see.

The creators claim that Oura is the best device on the market for analyzing your sleep and that it also estimates your readiness for everything that is awaiting you in the day ahead. Unfortunately, this first claim is not entirely 100% true (Oura is still very inaccurate in the estimation of sleep phases compared to a polysomnogram, as shown for example in this study and my long-term comparison with Dreem; see below). But the second claim about a readiness assessment is true, which will surprise you at first, considering how much less accurate, relatively speaking, the sleep data is. But there is a reason for this: sleep is not the only thing that Oura thoroughly evaluates.

You can choose from various colors, surfaces and designs

The foundations for using Oura is a well-built, regularly updated, clear and robust smartphone app. It helps you to see the quality of your sleep, along with other measured variables, and to understand how this quality of sleep and other even long-term factors contribute to your feeling of being rested or having energy (or sleepiness and fatigue). In addition, enthusiasts can subscribe to Oura Cloud on the web, where the data can be viewed on a larger screen and cross-referenced (“to look for correlations”) and play with them. After the last update, the possibility of meditating with Oura called Moment was also added, during which a person’s heart rate and heart rate variability (HRV) are evaluated.

The options for comparing different types of data in Oura Cloud are excellent

Let’s take a look at what Oura uses to construct its estimates and recommendations. The sleep evaluation starts with biometric sensors: an accelerometer (for motion), a pulse oximeter (for the pulse rate and HRV which is derived thereafter) and a thermometer (using the surface temperature of the finger) — all of which are active all night and are simply activated on their own. The ring appears to be equipped with an exceptionally accurate heart rate sensor, which then allows you to work with the exact HRV value (which is active even during meditation). Based on a sampling from the selected data during the night, Oura estimates, after you wake up and connect with the app, from when until when you really slept as well as the duration and order of sleep phases (light, REM, deep sleep and wake) from when you actually fell asleep until you woke up. Lastly, it calculates your sleep score on a scale from 1 to 100.

But that’s not all: the sleep score is really just a piece of the puzzle. It is itself a function of multiple values — the length of sleep, the length and regularity of sleep, the degree of restlessness, the ratio of sleep phases. The readiness score, which shows how you have managed to regenerate at night, depends not only on this one night, but also on other cumulative variables: your long-term condition and your level of activity during the last few days before the night that has just passed. Therefore, in evaluating your “day-to-day readiness”, in addition to the sleep score Oura also looks at other data available to it from its sensors or collected through the Health app (iOS) or the Google Fit app (Android).

As a result of this, I have been able to discover, for example, that minor differences in length (or phases) of sleep are not so crucial to my feeling of being rested (though differences of more than half an hour already do according to Oura). Above all, I have learned that the progress and value of my heart rate during each night is what matters more. It is not by accident that Oura evaluates the individual’s average value per night compared to their long-term average, as well as the lowest nighttime value of one’s heart rate and the relative time during which the minimum occurred at night. (The later you reach this minimum heart rate, the worse it is, because that means that until that moment the body was “dealing with something” instead of being able to rest.) In addition, it measures HRV for the entire night, a variable sensitive to the accuracy of the sensor (which is available from Oura, though). It is true that the lower the HRV (heart rate variability), the harder it probably is to find a balance between the sympathetic and parasympathetic systems, and the more likely you are either stressed or something else is wrong. In this case, regeneration is also unlikely to ideally take place and the reduced nighttime HRV will negatively show up in your daily readiness (but the HRV calculation is surprisingly not included in your Readiness score calculation).

But this still is not everything. To top it all off, Oura follows long-term trends in all respects, especially your physical activity over the last few days (if you do not wear Oura during the day and do sports with another sports monitoring device, it will read the values through your systemic Health/Google Fit app). For example, if you have been overdoing it over the last few days, or if you have a significant sleep deficit, it is clear that you probably won’t feel very “ready” after a bad night. Conversely, one worse night after a few days of balanced activity and sleep will be less of an issue in terms of your daily readiness. You also lose imaginary points towards the estimate of your regeneration if your nighttime temperature shows a significant change from your long-term average temperature (Oura only shows the relative deviation from your average, not an absolute value because this would be irrelevant). In this case, of if there are other deviations and anomalies, the Oura app will warn you that something is happening and anticipate that you will have a greater feeling of fatigue.

On the left, one night’s sleep along with the sleep score, and on the right a readiness score

The calculations and algorithms result in the readiness score. In the morning you look at it and find out how you are going to feel. 97? This is going to be a wonderful day, off to have a quick run and to start writing an article! 75? Oh, damn, I’m going to take it easy today, I’ll go out with the dog and go to bed early. All of this may sound like science fiction, but it works. No, don’t ask. Naturally, I verified this blindly. For a few days in a row, I didn’t look at the values in the morning, and wrote down my feeling of the day roughly in the usual Oura parameters. Then I compared both of them in the evening. The result? There was a subjective deviation of about 10%.


  • Oura is technically one of the most reliable devices I’ve ever seen. I would compare it with AirPods from Apple, for example, in that it “just works.” The same applies to the app, synchronization, updates, etc.; during eight months of daily use, there was only one data transfer error; this alone already places it among the best devices on the market.
  • The ring lasts about 2–3 days/night on one charge, which is pretty incredible. The charging is wireless (with a special charger) and the creators recommend that it should stay in the charger in various moments throughout the day — for instance while you are sitting at the computer, you can let it sit in the charger rather than letting it completely discharge and then needing to charge it for a long period of time and without breaks.
  • Due to its size, there is no issue with taking it with you on a business trip, you just simply wear it.
  • In addition, Oura looks really good, is comfortable to wear, it does not get in the way and is as tough as jewelry from Kryptonite (I don’t even take it off when I wash my hands, a normal ring would not be able to take this much abuse) — there isn’t even a scratch on it. Of course, it helps that I wear it on my left hand (I’m right-handed) and when I lift weights or something else heavy, I take Oura off.
  • When it comes to viewing, tracking and comparing trends in measured values, Oura is by far the best of the well-known devices (including Dreem) for me, both in terms of the app and its online applications.
  • It is quite reliable as an estimate on how you are going to feel and how you should arrange your day — it has helped me to organize my day and adjust or move trainings over time.
  • Oura does not light up or blink at night (as it measures by an infrared sensor). It does not vibrate either.


  • Unfortunately, when it comes to estimating sleep phases, Oura still has a long way to go. It has no EEG, and although it does its best, its error rate is quite high in my measurements (I give a very detailed comparison with Dreem at the end of the article). Most often, it over-calculated my amount of REM and deep sleep (at the expense of when I was only lightly sleeping) and not by a little: the values ​​differed from Dreem by up to two hours extra for REM and + 1 hour and 20 minutes for deep sleep. This is quite a disappointment to me, because if you want to track the development of REM length and deep sleep, you can’t take the Oura daily results very seriously. According to the aforementioned Oura study, it knows quite well that you are asleep, but it often labels things as sleep and other phases when you are (still/already) awake. According to the study, it will mistakenly think you are awake during your real sleep only 4% of the time, as opposed to when you are actually awake (so-called sensitivity is 96%). Unfortunately, in 48% of the time units when you’re not moving, it mistakenly thinks that you are asleep, even if this is not the case (so-called specificity). This is more or less confirmed by my findings. PS: Fortunately, the start and end times of sleep can be adjusted manually the next day, depending on what you can recall about the night and the morning. However, the types of phases cannot be edited, nor would it make sense to do so.

The Verdict: Oura is a hi-tech jewel in the literal and less literal sense. It is a small miracle and we will surely see further improvements and functions, perhaps in the next generation, because its Finnish creators clearly do not sleep. Its greatest benefit for me was that it taught me to perceive the long-term complex relationship between sleep, physical activity, nighttime heart rate and rest. Unfortunately, as its weakness I consider the limitations of its biometrics and the resulting inaccuracies in the recognition of sleep phases. So in the end, it is probably closer to reality than other similar gadgets of a similar size (i.e. bracelets and rings), but it is unable to be on par with gadgets containing an EEG device.

Go2Sleep — Only for Sleep Apnea

Let’s move on to another device — SleepOn Go2Sleep. It is a South Korean startup that is also the result of crowdfunding. It is a measuring capsule inserted into a silicone case that slides onto the finger before sleep; but it cannot be called a “ring” — it looks more like a medical device and in a way it is also trying to be one. It uses similar biometric features as Oura for measuring: an accelerometer (for motion sensing) and a pulse oximeter (for heart rate). In addition to these traditional bits of data, it also measures SpO2, i.e. oxygen saturation. This figure is one of the key factors in the laboratory evaluation of whether you suffer from sleep apnea or hypopnea, a serious chronic disease that can lead to death, and not just your own (in more than one way). As for the app, it has a number of great features that are hard to find elsewhere.

My monthly sleep report last year, when I was newly using Go2Sleep on a daily basis


  • Go2Sleep is a lightweight and portable device like Oura, which provides a unique ability to continuously measure blood oxygen saturation; if — you snore; are overweight; your partner catches you occasionally catching your breath at night without waking up; wake up tired or fall asleep during the day or even while driving… you should look for a doctor right away! :( But probably you are hesitating, since you don’t want to be considered a hypochondriac, are afraid to make “such a move” and most of all, you worry they may find “something”. You wish. Well, this relatively effortless and multi-day screening of SpO2 values with Go2Sleep could prove to be the final straw that breaks the camel’s back and gets you go to a specialist [2] — or you will use it to get a snoring partner there.
  • The Go2Sleep can even deliberately wake you up by vibration when it detects that you are choking; I have not tried this — based on the evaluations and observations ​​from the device I probably don’t suffer from sleep apnea, so I couldn’t see this for myself.
  • The Go2Sleep app has re-invented itself twice during its existence and always for the better — the creators are not idle; it has features such as labeling “sleep circumstances” (you put a “sex” label one evening, “drugs” the other day, “rock n’ roll” a third, and later you correlate your long-term sleep score — by the way, can you guess which is the best for sleeping? :)
  • The app can even link multiple accounts (e.g. in the family) and make comparisons, of course, if others also have Go2Sleep; the possibility of online monitoring and warning (for example, a baby for whom a loss of oxygenation may be a sign of a life-threatening issue) — these are all unique functionalities.
  • The app allows very detailed “minute-by-minute” nightly recording outputs.
A detailed sleep report from one night — I wish that the data could be trusted more


  • You may be thinking that this all sounds pretty good, but if you poor people have already read the previously long part of this article, then you already know that this comes with a catch — and that is primarily (in) accuracy of measurement and the poor quality of its sleep evaluation. Unfortunately, Go2Sleep is not very good in comparison with Dreem, and it is worse than Oura.
  • While the consistency of the Oura or Dreem results is high, in the case of Go2Sleep it seems as if all of a sudden it was measuring the whole night with some other sensor or working with a different algorithm — so the result differs radically without the night itself indicating subjectively any kind of anomaly; I think it is a design error that I will mention below.
  • Additional doubts have arisen from the statement of the creators of the Go2Sleep claiming they can accurately determine the so-called AHI index; nowhere did I find a study to prove its credibly; the necessary data that a doctor needs to confirm that you are suffering from this disease is more than the value of SpO2 itself. What also gets measured is whether in fact you really (and for more than 10 seconds uninterrupted) do not breathe according to chest rises, muscle tension, etc. (see picture below)); only out of these factors can AHI then be derived. According to Wikipedia sources, though, SpO2 for home screening is enough for early detection, and Go2Sleep can actually act as an early warning device — if your SpO2 regularly falls below the recommended values ​​overnight, do something about it!
A polysomnogram indicating the typical symptoms of a sleep apnea episode. Source: Wavedaker NV et al. Best Practice of Medicine. Sept. 1999.
  • The application is beautiful but unreliable; the data transfer is equally unreliable: it has happened to me more than once that the data wasn’t transferred or that only a portion of it was transmitted (for example, there were average values, but the night chart was empty, etc.), which is really frustrating when you are playing around with the data to this extent and are following every twist it takes.
  • The sensor on your finger is continuously lit all night (or more precisely: it is blinking, red and green); the creators have confirmed to me that it is “not biologically toxic” in relation to circadian rhythms, but if you see green light, I don’t believe it much (see for example my article on a similar problem with the Apple Watch): the constant flicker itself is quite distracting for you and others around you. It helps if you keep your hand closed, but you won’t always be able to take care of this.
  • When it comes to the design solution: a silicone case is not a good idea. The capsule tends to spin inside of it, and this risk increases with time; after two months, as the silicone softens the capsule may even fall out; and just turning the capsule leads to erroneous results, as I have repeatedly found; in terms of the points Go2Sleep’s design solution gets, it got one star out of five, if this makes it easier for you to understand.
SleepOn’s Go2Sleep in its charger, which uses two contact points, not induction

The Verdict: Go2Sleep is quite interesting, but technically not a completely worked out sleep tracker that comes with the typical errors of this type of device. Perhaps I would only consider it in cases where there is a suspicion related to apnea as I mentioned. Its ability to measure SpO2 values will be quite accurate, more accurate than its ability to measure other sleep values.

Withings Sleep — Waiting Faithfully in Bed

Do you remember Beddit (it too came into existence through crowdfunding)? I praised it a few years ago — it was an innovative solution for its time. Beddit followed one’s sleep through a sensor placed under the bed sheet at chest level. After a few years, Beddit began to lag behind the others, and eventually Apple acquired it from its creators. An Apple Watch still does not measure sleep, so it is hard to judge what the acqui-hired team of experts in Cupertino is doing … :)

Meanwhile, other creators entered the market to try to fill the space under the sheet instead of Beddit. One of them is Withings, a company of no less a troubled history. Their Withings Sleep Tracking Mat prides itself on detecting breathing disorders during sleep. It inputs various biometrics such as the vibrations and movements that the human body makes during sleep. These vibrations are apparently able to be broken down by the device into prime factors: heartbeats, inhalation, exhalation, etc. In addition, the device has a microphone that monitors snoring. This is the only device in my test group that has to be permanently plugged in.


  • Withings Sleep Tracking Mat doesn’t have to be worn, nor does it need to be recharged — you simply lie down and get up in the morning without having to do anything, it waits for you to come back, the data is synchronized via Wi-Fi, etc.
  • Assuming we can trust Withings, the device can “estimate the risk of suffering from apnea based on a unique algorithm capable of recognizing breathing irregularities.” Unfortunately, I cannot judge this and I haven’t found any studies about this. As was mentioned previously, using only one variable is not sufficient for diagnosing the disease, but for example a stop in breathing and an accelerated pulse may probably suggest something similar to Go2Sleep decreasing Sp02;
  • The design of the device is nice, sturdy, and does not get in the way in bed.
  • Health Mate is quite a minimalist app at first glance, but if you want, you can comfortably get acquainted with the most detailed data and a broader timeline; the app is common to other devices with Withings that you have interconnected, such as a weight or pressure gauge, etc.


  • I consider the greatest con, a shortcoming that needs to be commented on directly, to be its inability to know that I am still sleeping; in other words, almost 2–3 times a week, the measurement will be cut off at about four or five o’clock, even though I just stepped out to go to the bathroom (which I do daily in the morning) and then no longer measured. In the morning, you can additionally adjust the wake-up time, but the “fill in the remaining time by sleep” option is not enough; this will fill all the time when you were awake at night with “sleep” of unknown quality; this made me very disappointed and I consider it a deal breaker (until it gets corrected); I would like to express one more regret: I can understand all the wearable devices that may not realize you are lying in bed, since this is actually quite hard — but I don’t understand how these clever little sheets in the bed, where you lay the weight of your body, cannot.
  • The accuracy of deep sleep measurements compared to Dreem is poor (over-average by 57 minutes ± 37 minutes), but REM is measured much better than Oura (-9 ± 34 minutes); but I remind you that this data is burdened with the error caused by having to calculate the missing parts of the night (the calculation algorithm is not known to me).
  • By nature, it is not convenient to take your “bedding” with you on the road, it is not simply a travel device; it has a cable and a transformer and the installation is just a total hassle and you always run the risk of shifting it and thus changing the consistency of the measurement (and I am not even counting the fact that the hotel staff will probably find it when you leave the hotel).
  • Although Withings Health Mate includes Sleep Coach, Withings Sleep Coaching is not as sophisticated as Dreem or Oura; it is more generic and not very tailored to your sleep and level of activity as are Oura’s personalized analysis or Dreem’s immaculately done typological questionnaires.

The Verdict: Perhaps I am unfairly judging Withings Sleep Tracking Mat, but I hesitate to recommend buying it: it got it completely wrong when it came to my deep sleep measurements and it short-changed me when it came to my data (both when I slept and it wasn’t measuring as well as the data that won’t be measured if you don’t take it with you on a business trip). However, its analysis of sleep problems can be beneficial for similar reasons as Go2Sleep. Unfortunately, I am not able to compare this parameter for both devices.

Day after Day: What Does the Data Tell Us?

Now, here comes the most interesting part (and I would frankly say the most sophisticated part): how do these individual sleep trackers fare when we put them next to each other and look at the measured data? I have to say that some of the partial comparisons pleasantly surprised me, but overall the theses outlined in the introduction have been confirmed: the distinction of the phases with secondary biometrics (at least according to me, Tomáš Baránek (46) through May 2019 [3]) is still very inaccurate.

But little by little, I wanted to look at two things:

  1. I wanted to compare the alleged duration of REM, deep sleep and light sleep for about 30 nights and put Dreem, Oura, Withings Sleep Tracking Mat and Go2Sleep data side by side. Unfortunately, the Go2Sleep statistics for the whole month were ruined after the Go2Sleep app update (which is typical and a warning about this device); Therefore, Go2Sleep is missing in the long-term comparison, I won’t be taking measurements from it for another month.
  2. I also wanted to compare the resulting hypnograms from a one-night’s “minute-by-minute” measurement, especially with Dreem, and to look at the typical errors in phase interpretation (following the findings from point 1). Now it was Withings Sleep Tracking Mat that repeatedly failed, so measuring with this gadget is marked by the fact that half of the data is missing for both nights when I wanted to do the measurements. [4]

You may be disappointed that I have not compared the above reviewed devices with the more common and cheaper sleeper trackers (sleep apps, watches and bracelets, etc.). Let me answer this simply: I didn’t want to waste my time on something that is, from the beginning and in essence, doomed to the worst possible results. All of these devices work on the basis of one (motion / sound), preferably two (heart rate) metrics, none of which are reliable, and is de facto out of the question that it could be compared to devices like Oura or even Dreem. So, I just didn’t go there except when I was tempted to try out at least the popular Sleep Cycle app (to estimate the phases it uses the microphone) and Pillow (which uses motion and the heart rate on the Apple Watch), but the results were so bad that I really had no reason to continue.

The Comparison of Average Values from May

First of all, I must point out that I treated Dreem as the standard and the default “right” value for comparing other meters. According to the above-mentioned studies, its accuracy is both very high and approaching a polysomnogram (and that is to such an extent that even the differences between a PSG interpretation by different experts are higher than the differences between Dreem and a PSG). For the anecdotal proof of the accuracy of the minute-to-second estimation, I can also take the fact that I have never noticed any relatively strong sound stimulation from Dreem, which I have completed about 9000 up to today (UPDATED on 7/12/19 — now at 18000 and still nothing :). That would match the fact that it differentiates the course of each phase exactly up to the second, because if it were wrong, it would wake me.

Therefore, compared to the other measurement techniques used by me, the Dreem analysis is most likely to be the closest to a PSG and to an interpretation by an expert.

What and how did I compare? I exported the phase lengths for each night and pasted them into a large comparison table. One night, the Dreem completely fell off my head, so these measurements weren’t recorded, so this night is missing completely. When I wasn’t sleeping on the bed with the Withings Sleep Tracking Mat, I still included the data from the night (just without any Withings data). Thus, the data intersection remained 21 nights including Withings (and 27 nights without it). Below you can see a graph of the comparison of the measured lengths of NREM (deep sleep). Dreem is Red, Oura is Blue, Withings is Green.

If you look at the dispersion of the plotted values of Oura and Withings in relation to Dreem, you will notice unpleasant moments: at first glance, they differ by tens of minutes. Even worse is that even the overall trends do not fit together. There are a shorter series of several days when, according to Dreem, my deep sleep was getting shorter / longer, but Oura or Withings show exactly the opposite trend. The same applies to REM, where the differences are even more striking:

One then has to ask whether the measures (more movement, no alcohol, less light in the evening, etc.) that we usually compare with the measured sleep values make sense to compare against these meters on a daily basis. Neither Oura nor Withings really convinced me that it made sense. (Not speaking of the cheaper solutions…)

What about longer-term trends? Here is where it gets interesting. At first glance, the sample of my 27 days long-term trendlines do not fit much (these are the color lines in the charts). While Dreem determined that my REM sleep had improved significantly in May, Oura, on the contrary, suggests a decline and Withings stagnation. When it comes to deep sleep, Withings goes even in the opposite direction to Dreem. But Oura showed something truly remarkable. It was fairly accurate at evaluating deep sleep during the first half of the month, but then it started to diverge from Dreem. While Oura claimed that there was an increase in deep sleep, Dreem said that it had remained more or less the same. I have a hypothesis about this: from the middle of the month, the effect of stimulation to deepen deep sleep began to increase. When it is even deeper, maybe the brain and body will relax and even handle the next phase of the night? At least this is indicated by data from Dreem, where the length of REM begins to grow around 10 May. Why did this affect Oura’s estimate of deep sleep length, we can only speculate; Perhaps it is precisely with the Oura’s heuristics, which also considers the dividing phases as “stimulated” deeper deep sleep.

I had sincerely hoped that my favorite Oura would offer me more convincing results in the long term, but this was not (yet) confirmed, but rather refuted; In the 27 days measured, a reliable data sample is not entirely reliable for assessing the quality of long-term trend measurements in the sleep phase length.

Check out the summarizing (averaging) table above. I tried to evaluate the differences in measurements statistically. The result is that Oura overestimated my deep sleep per day by an average of 17 minutes (with a standard deviation of ± 30 minutes — so the result is mostly from -13 to 47 minutes), the maximum is +1 hour and 18 minutes. Oura added an average of 38 minutes a day (± 36 minutes) to REM with a maximum of 1 hour 54 minutes. Both poor phase time estimates tend to be at the expense of the light sleep length (which Oura underestimated on average by 59 minutes ± 37 minutes). So the range of values is quite wide: even 25% of the time of the night.

The Dreem was the only one that could reliably know when I really fall asleep and when I am awake (both at night and in the morning). I know it simply because I remember it well almost every day :).

Oura had a problem with the morning phase, when it considered quiet watchful sleep or light sleep REM sleep, which spoils its average in the area of REM / light sleep. So for all sleep trackers except for Dreem, you have to manually adjust the beginnings and endings of the nights almost every day, which I did at the time of the measurement, so paradoxically, the result distorts the reality for the better for all devices except Dreem when it comes to the exact estimate of the total sleep time.

The Withings Sleep Tracking Mat is also not doing well in terms of its estimates for REM / deep sleep / light sleep. Moreover, its data is burdened by the mentioned tendency of cutting off of the measurements late at night. However, I did not simply reject the resulting data and compared it again: the device added an average of 57 minutes to the deep sleep phase (± 37 minutes), reduced the REM phase (-9 minutes ± 34 minutes), also reduced the light sleep phase (- 14 ± 42 minutes).

My conclusion, therefore, is that Withings and even my favorite Oura are not reliable for accurate phase evaluation. Oura has the smallest standard deviation from the average error (its dispersion of values from Dreem is less off the charts for each phase), so it is a little ahead of Withings and probably others. Withings, on the other hand, was a little more consistent and accurate when it came to REM sleep results. I want to believe in the future that if given a period of more months there might be at least a basic agreement with the reality of the overall trend in the length of “high-quality” REM + deep sleep (averaged e.g. after a few weeks).

Detailed Comparison of a Randomly Selected Night

The second examination will be an analysis of a particular night when I compared Dreem, Oura, Withings and even the unfortunate Go2Sleep (there is only a screenshot of its measurements, since more detailed data were not preserved, but that’s enough). To emphasize the differences in the estimation of the onset and duration of the phases, I distorted the applications hypnograms to fit the Dreem chart exactly and made them more transparent. So Dreem measurements are the horizontal bars showing through all three graphs.

During the measured night Oura well estimated the onset of sleep and showed a reasonably good match in the first hour of sleep, then the phases started differing, though. By about three o’clock in the morning, it thought I had been up 20 minutes (and yet it was REM), and then it considered about 50 minutes to be deep sleep, although it was light sleep. Subsequently, I was for about 40 minutes intermittently going between light sleep / waking up, but Oura simply threw it together in one bag: it marked it as me being awake all the time. The next two or three hours it was measuring more or less fine but before awakening it showed a half an hour of my light sleep as REM:

As far as Withings Sleep is concerned, we have already said that it had trouble knowing when I was in bed. Withings in the first phase of the night did not notice that there were two times when I was completely awake (and that was for a few minutes, I even remembered it then) and then around 4am it gave up on measuring altogether (because I was up for longer that time, went to use the bathroom, then I stayed in bed trying to sleep):

Finally, there is SleepOn’s Go2Sleep. As you can see at first glance, in the first part of the night, the deep sleep and REM phases tend to be mistaken for light sleep, then it wrongly assumes almost an hour of light sleep to be REM, followed by a wrong estimate of the length of waking up time, almost two hours of light sleep are marked as REM and it claims the last phase of the morning to be solely light sleep, although it was a steady mix of waking up + REM + light sleep:

Some Good Advice at the End: It’s Better to Do Things Properly than Poorly

It has to be said that despite all the foreign words and references to the sources I use in the article, my comparison is still quite amateurish and loaded with a lot of ignorance and flaws. I had relatively little data, the measurements were done on a single subject (me, a 46-year-old, white man with sleep problems, taking low doses of SSRI) over a short period of time. Therefore, take these conclusions with a grain of salt and continue to use your brain:

  • Dreem is, in my opinion, the best device at this point and if you are having trouble sleeping, or wanting to deal with it seriously, it pays off to invest in it; it is not a “fancy” device that will glitter on your finger also during the day, nor will it measure your level of physical activity, but it will really help you with your sleep. The data from it is accurate enough for the given need, the coaching about sleep hygiene is great and the stimulations are likely to be restorative when it comes to sleep (and this quality will be most likely even more enhanced by the data collection from more users). If they add better tracking trends over time (such as on the web) and fine-tune different details, it will be an absolute blockbuster.
  • Oura is not as good at estimating sleep phases as I thought, but it was actually a mistake based on my lack of education about it — it cannot be any better. Despite that and although with a long way of catching up with Dreem, it is the second best overall and actually everything else (except for the phase precision) proved to be good about it: from its ingenious physical execution through the application to its way of estimating your readiness for the day ahead (which is clearly not solely based on phase proportions but also on other factors). If you are rather younger and curious, have no major problems with sleep, or are not looking at the stages as a key variable, but are rather focused on your overall length of sleep and performance, Oura is a solid and very comfortable helper.
  • Withings Sleep Tracking Mat and Go2Sleep are both devices that focus on breathing difficulties and apnea screening. I would take that as the starting point for making comparisons as by doing so the comparisons will do them the least harm; yet I cannot really judge this service of theirs well; the reliability of both devices under real conditions of use is worse than that of above mentioned favorites (both of these devices have a problem with data processing). They aren’t so comfortable either, because Withings is under the bedding, the Go2Sleep flashes and over time it can fall out of its silicone case and lasts only one night for one charge. Yet their contribution to warning of a serious health problem (sleep apnea) may be significant, and if you suspect you are suffering from this, I would consider trying something along these lines.
  • What about all the sleep apps for mobile phones (even with Apple Watch)? All you can count on them reliably measuring is your level of physical activity at night (tossing), snoring, progression and heart rate average; if you add a manually corrected start and end of sleep, you have a data set for further processing. But do not engage in them with a hope of exploring the alleged phase differences.

If you have read up to here, I can only thank you for your great patience and faith. Please write your comments, experiences and questions in the comments section below the article, not via social media networks, because I would like to leave this discussion easily accessible to all readers. So, what do you think? Are you going to Dreem? :)


Tip: If you understand Czech, you can read the original Czech version or read more articles on my blog lifehacky.cz and subscribe to my lifehack newsletter and prepare to always learn something new. I almost forgot — I also give gifts to subscribers. You can unsubscribe at any time. On principle, I will never send spam. There is even an archive of older e-mails.



  1. Deep Learning Algorithm Dreem is being used by five sleep scientists to learn more, here they are being presented with pre-anonymized records of biometrics from different sleepers and they interpret these polysomnograms like they do in the lab. The algorithm constantly teaches itself. For more details: https://support.dreem.com/hc/en-us/articles/360019948151-Classification-of-sleep-stages
  2. I couldn’t find confirmed reports on the accuracy of measuring SpO2 using Go2Sleep anywhere, but I did find some general studies on the method of using SpO2 measurement based on pulse oximetry: the accuracy is close to 98% compared to the precision of an ABG analysis (for interest: the highest accuracy is in ear lobe measurements). See, e.g., https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5905124/
  3. This is a very important footnote. As I went through the various studies, it turns out that with increasing age, the accuracy of measurements is different with different sleep trackers compared to polysomnograms. This is likely to correspond with deteriorating sleep (increased fragmentation, restlessness and vice versa, which shortens the entire measured interval), resulting in much more inaccuracies.
  4. While I could measure another 30 nights again with Go2Sleep and I could also choose the night where Withings Sleep failed: however, because of the pretty typical behavior of both devices, which I consider to be a deal breaker, I simply let this mistake be reflected in the methodology.


I would like to acknowledge the efforts of Scott Hudson, who translated my original article from Czech to English with great care and sensitivity.

Tomáš Baránek

Written by

Writer, publisher, father and son. And .

More From Medium

Also tagged Gadgets

Related reads

Welcome to a place where words matter. On Medium, smart voices and original ideas take center stage - with no ads in sight. Watch
Follow all the topics you care about, and we’ll deliver the best stories for you to your homepage and inbox. Explore
Get unlimited access to the best stories on Medium — and support writers while you’re at it. Just $5/month. Upgrade