I am a little late to the party and have a few thoughts based on my experiences with a few mmWave sensors this past year. I have a number of projects which you can follow in the HA Communities.
mmWave Presence Detection
This is a pretty exciting project and there is a lot to discuss, let’s get to it
mmWave - Where I see us today
As of June 2022 I have tried four (4) different mmWave sensor modules with varying capabilities. They have taught me a few things which have been echo’d earlier in the thread. Please consider me a layman in this area; one with a little experience of the basics.
So what are mmWave basics worth considering?
- the antenna array which will influence the detection area and impact physical design (size/placement)
- the PCB/enclosure design which will influence detection direction (going backwards is possible)
- the MCU, F/W, configuration parameters which will influence the detection trigger time (a lot more on this below)
- the FTT analysis which will influence featureset/configuration-possibilities
I suspect that with the proposed placement behind the paddle there is sufficient room for most [recessed] antenna design. The biggest contraint is already mentioned, it likely means +/-170° at best!. This needs significant testing in order to understand the relationship between detection distance/sensitity and end-of-range/angle detection/sensitity. mmWave is just a radio wave after all and it drops in strength at the edges of its dispersion pattern. There is a tradeoff in antenna design/gain/distance/angle/etc. I will leave that to the PhD design experts with the understanding that you can’t have everything without paying for it.
What I could envision are two arrays protruding (30°?) from the switch (ala PIR lens) for greater coverage, I am not clear there is a requirement or if there is acceptance factor for this.
This seems like a requirement to dictate to the sensor module manufacturer in order to ensure no rear detection is possible. I have a SeeedStudio 24Ghz mmWave sensor module that does this and requires a metal box to ensure forward facing detection.
MCU, F/W, Cost
This will honestly be the biggest contributor to “mmWave performance.” mmWave in itself is just radar (doppler/FMCW/CW/etc) and the “interpretation” of those radioforms is key. The faster the MCU, the more cost and not necessarily the better performance as firmware influences this. Just like an application on your PC, the greater the workload the slower it runs. This leads to the next section…
This is the app the mmWave MCU runs. If I were to categorize these based on my experience from simple to complex;
Simple. The DFRobot or equivalent Leapmmw sensors that perform the most simplified movement-based analysis. Human micro-kinesis at best.
– Pro: fastest response time ~25ms
– Con: fans, AC’s, airflow wiggling a curtain are all potential triggers.
– Risk: without a PIR this a challenge based on my limited experience.
Medium Complex: This is your SeeedStudio “Human/Heart/Fall” or Aqara FP1 type module that performs “interpretive analysis” on the mmWave FTT target scatterplot.
– Pro: Can be programmed to recognize “human’ish movement patterns” which may inherently ignore object-based movements.
– Con: 2022 versions are slow. >1 sec trigger time and as the Aqara demonstrates can get much worse.
– Risk: you are at the whim of the algorithm for detection resilience.
Complex: This is your RFBeeam KLD-2 or KLD-7 type sensor that do all sorts of amazing things.
– Con: first up, cost!
– Pro: awesome’sauce
– Examples: The KLD-2 detects approaching/receding as well as direction in <800ms and can be configured below 400ms. The KLD-7 is not intended for this application.
The Unknown: "what about the [insert hundred other AliExpress modules here]
– Risk: you need to test it. Nothing can be assumed without validating a comprehensive datasheet. No datasheet, don’t bother.
Requirements & My Vote
Design decisions are needed in order to determine a suitable module for this application. Quite frankly, there is no perfect answer. My thoughts are;
– <100ms response time. There is nothing worse that slow lighting and the number one reason to toss out a sensor.
– A fast sensor that cannot be tuned to a wide variety of installation situations will just as quickly be binned. What you lose in wall/switch placement flexibility you must compensate for with detection parameters. Sensitivity is not enough, without distance at minimum you will quickly accumulate those single star reviews. Direction/angle/speed all add to the cost/detection-performance
– Cannot penetrate walls. This is a dealbreaker. I didn’t believe this at first until I owned a SeeedStudio.
– Wall contruction must be defined.
My 2022 Vote
A non-interpretive-based mmWave sensor module/MCU that is fast and permits at minimum distance and sensitivity adjustments. Combined with a [good] PIR it would eliminate the majority of object-based [non-heat generating] false positives. Having been quoted $15 (China-direct) per sensor module in single digit quantities, I suspect this is closer to $10-12 for the mmWave in quantity. This gives you $5-10 for a good PIR? If the $150 version has anything less than a Panasonic EKMx PaPIR I will be cry /s
My 2023 Vote
An interpretive-based mmWave sensor module/MCU that ditches the PIR for waveform analysis to detect human-type motion in the target detection period [<100ms].
This thread is long with a lot of great suggestions in a wide variety of areas. Is there a tabulated breakdown of the final direction?
- Physical design
– button placement
– sensor placement
– artistic look
- Conceptual Design
– what is my target response time?
– what is my target detection area?
– what are my configuration parameters? We have discussed a few, this should be investigated further; eg channel selection (interference avoidance)
- Logical Design
– spectrum (5.8/24/60) - impacts gain/distance/interference. My vote: avoid 5.8, 24 for distance, 60 might be immature this year?
Conclusion & Thoughts
In my journey of deploying no less than seven (7) mmWave sensors in 900sq feet I have to conclude; done right, this will be a game changer. A non-interpretive [DFRobot] sensor is great for rooms where object-based detection is of little concern. Yet I suspect this is a small percentage of deployments. For all the rest, if you don’t chose a human-interpretive or PIR+mmWave combination you will experience a significant slope in excitement-factor and then a mirrored decline. I have had a dual-mmWave+PIR combo in my kitchen for months and it is reliable and what you want to deploy without fast interpretive option. Given the added placement contraints of a wall switch, anything less will be extremely challenging to gain long-term acceptance.