Note how this might be "Wifi 6", but it's still 2.4Ghz only and as far as I can see still single antenna (making most mimo functionality useless). Target Wake Time looks like the only feature that might be useful (reduces battery consumption substantially), but few applications can make use of that yet because your product still needs to work if the user has an old router which doesn't support it.
Other ESP32's in the range don't get more than about 20Mbps throughput under ideal conditions, so it probably isn't exactly a beast in the wifi performance department.
In fact, I wouldn't be surprised if this isn't the same silicon with an upgraded firmware to enable this new functionality.
Shouldn't the fact that it's WiFi 6 mean it uses much less air time? I thought WiFi 6 was designed for this scenario with many clients. 2.4GHz is a bonus for low bandwidth iot devices where range trumps throughout.
Yes, 802.11ax added all the technology required to allow subchannel slicing in both the time and frequency domains.
Lets say you had a 5ghz 8x8 160mhz AP (maximum allowable by spec): if all you had were 1x1 20mhz clients, you could theoretically have 64 different clients concurrently communicating with the AP without any real problems.
2.4ghz does not have 160mhz channels, but allows up to 40mhz. Ignoring the fact that 2.4ghz is incredibly congested, and you will never achieve peak performance, you can still achieve a theoretical maximum (again, with 1x1 20mhz clients) of 16 concurrent clients.
That said, I'm more interested in 802.11be: MIMO across bands. Instead of hard disabling 2.4ghz (which I 100% recommend all people do: this cures all Wifi ills; and if you have a building that eats 5ghz, either move or spam APs), the AP will dynamically use 2.4ghz as a possible MIMO band.
Theoretical maximum configuration of 802.11be is 40+160+320 (one 40mhz 2.4ghz channel, one 160mhz 5ghz channel, 1 320mhz 6ghz channel), and 802.11be devices will simultaneously communicate across all of them; maximum MIMO configuration has also been increased to 16.
In other words, the future is an AP that looks like a hedgehog, and has 10gbit port.
> if you have a building that eats 5ghz, either move or spam APs
The insanity of moving just to get better wifi reception aside, do you mean move to North America where houses are built out of cardboard and sticks? Most other countries build walls out of solid materials and 5 GHz isn't getting through more than maybe one of those.
As for spamming APs, I really wish it were that simple. Apple devices cling to their AP until speeds drop to the dialup days and other platforms aren't much better. How did we figure out GSM roaming in the 90s but WiFi roaming is still a disaster 30 years later?
If you've set up many APs already, and find client devices sticking to them too strongly, this is a sign to turn down transmit power on the AP in question. Most ship set to "Max" which is terrible for SNR and spatial separation. Turning down transmit power will cause the client to switch sooner, non-intuitively providing more bandwidth due to better utilization of spatial separation.
I've seen that work well in open areas where APs overlap significantly, but in places with walls where you walk from one bubble to another, it still usually takes to long for a device to figure out it should switch and then actually do it. I've never had a video call even stutter when moving between LTE cells, but on WiFi, calls sometimes drop outright, even when walking down a long hallway with many APs.
The mere fact there is a single knob for 'transmit power', rather than the transmit power being automatically determined by the clients location relative to the AP, other AP's, and noise sources tells you a lot about the maturity of WiFi as a spec...
Deploying Ubiquiti's APs in managed (as opposed to standalone) mode makes this easy; even iPhones will transparently be passed from one AP to the next. I agree, though, the OSX/iOS roaming preference tuning is a bit batshit; they also refuse to drop your house's Wifi as you're driving away and want to switch to LTE ASAP.
Commodity/consumer APs aren't designed for handoff.
OH!!! Now I understand why people say 5Ghz doesn't go though walls!
I knew other countries had heavily walls but didn't make that connection.
I quite like our cardboard walls though, they are trivial to repair and upgrade, and they rarely get damaged in real life, except if a Real Tough Guy punches one.
As long as there isn't any other stuff on 2.4 GHz, it can indeed be beneficial.
But if there are any legacy 802.11 (especially pre-802.11n) or non-802.11 transmitters nearby, it can be catastrophic for throughput, both because you need protective measures when transmitting (so that older 802.11 versions can reliably detect your transmissions as occupying the carrier) and because some of them just don't cooperate well with 802.11 at all (e.g. Bluetooth, which will just happily transmit over 802.11 without any regard for CSMA/CA fairness at all).
The lack of 5Ghz in all their products is... surprising...
There are plenty of home users with 5Ghz only networks, which in turn means anyone designing an IoT device will get user complaints and returns from all these users.
>There are plenty of home users with 5Ghz only networks
really? is this a configuration that some ISPs are shipping in their routers, or how is this happening? i've never heard of anybody with a 5GHz only home network.
It's certainly somewhat common for more tech savvy folks to go out of their way to ensure their devices use 5Ghz when compatible.
But ISPs? Router manufaturers? No way in hell, why would they self-impose additional support burden by preventing the customer from connecting their 4-5yo old low-end laptop or their shiny new smart bulbs or their brand new Amazon Kindle?
That's an odd configuration, because having a 2.4 GHz network that all the shitty IoT devices can be on is one of the best remaining uses of that band.
2.4 is still the best choice for range and penetration (through walls etc.) if you don't need the speed of 5ghz. I'm honestly surprised if they're shipping 5ghz-only consumer access points.
> The lack of 5Ghz in all their products is... surprising...
Is it? Are there any competitors in the class have 5ghz support? Most of the radio MCUs I've seen are all 2.4ghz which includes thread/zigbee/wifi/etc.
To be fair, Espressif aren't the only ones. SiLabs announced the SiWx917 eons ago, and they finally appeared on Digi-Key like 3 weeks ago, but only in bulk, and the only dev board available is $200, and no Zephyr support, and this, and that...
> single antenna (making most mimo functionality useless)
Can't the AP still use MU-MIMO even for single-antenna clients? That's still a significant improvement every time more than one STA is active, especially in the super-busy 2.4 GHz band.
This is the type of device where 802.11ah (recently popular on the front page again) or LoRa is killer. Unless you need your ESP32 to push 10s of Mbps all the time you can instead use something which punches through multiple walls with ease, doesn't compete with high speed devices, and has power saving things like TWT years ahead of high speed Wi-Fi standards.
Yes, but (nearly) every potential buyer of IoT device has 2.4Ghz WiFi network at the location that said IoT will be deployed.
LoRa will require some kind of bridge device, and since we can't agree on anything - each vendor will have their own bridge device.
Alternatives like Amazon Sidewalk and Helium could work, but:
- Helium means users will pay for traffic to someone and use someone's internet connection. Doubt it will fly. It would be cool in a post-apocalyptic world, though.
- Amazon Sidewalk isn't universal and, IMO, immoral (enabled on devices by default without informing the owner of the device). Good luck downloading updates to your light bulbs at 80Kbps.
Just yesterday was looking at ESP32 custom board used to open/close gates which has super weak signal even when access point is only few meters apart... I found that there is empty place where external antenna connects - I think I can solder the connector and connect the antenna.
Is there any article out there that would help me do it? As I understand there needs to be some resistor that should be desoldered or existing link to on-board antenna cut.
That's about it. Antennas aren't anything but wire or a small sheet of metal, all you need is something to increase the amount of metal exposed to the signal.
This is a quick & dirty diy "solder on some more metal & an antenna" tutorial:
YMMV, but I have tried this several times and have found it very difficult without microsoldering equipment.
Some ESP variants have a built-in U.FL connector which makes things easier (e.g. AITHINKER ESP32-Cam and its many clones), but you still have to remove the 0Ω resistor, which I find very difficult with even the smallest tips on my Hakko iron.
Could be a skill issue on my end; should be much easier if you have a simple reflow/rework station.
The antennas are a commodity item. You can get them dirt cheap at your least favorite online retailer. The connector is most likely a uFL type. You could also ditch the connector and solder the antenna lead directly to the board. Outer grounding braid goes to ground, inner conductor is the signal.
I will start by saying I am not an embedded engineer. Given that I am but a consumer of about a dozen different IoT devices, I would love it if embedded wifi chipsets didn’t suck so much. I don’t care if it’s WiFi 4, just make them not suck.
Each manufacturer seems to have their own little quirks you need to work around: one device only works well on 2.4 channel 1, another will aggressively roam to other access points (even though it and the ap are stationary, and the closest ap has < -60dbm signal). Even the most relatively well behaved modules will disconnect and reconnect constantly (average connection time is ~10 minutes to an hour, mind you the device is constantly connected to mains power). God forbid you enable any 802.11 extensions on the radio servicing those devices, then you’re really in for a world of hurt.
Then when you complain to the manufacturer, you get super helpful advice like “factory reset your router” and “open the following 28 ports through your nat gateway to our device” (sooooo… and if I have two of your devices, what then?). It’s a total shitshow.
Further I think most of the people using ESPHome and the like are reflashing commercial products.
> the go-to chip for commercial applications?
For a pacemaker, a Mars rover or a video game console? What exactly do you think unifies “commercial application” that would allow narrowing down in such a way.
If you’re talking about low cost IoT, there are a handful of vendors including Beken, Realtek, SiLabs, ST, TI, nRF. For this super low cost WiFi/BLE consumer niche espressif is not the only player but they are dominant along with nRF and Beken.
One concern is supply chain and working with a Chinese company. If you build something that has to guarantee a 10 or 20 year supply some companies might balk at the ESP32. I think Espressif has jumped over into the "reputable" pool but it could still be a concern.
On the other hand when someone like TI prices their parts at like 30x the cost (e.g. ESP32-WROOM $3, TI CC3235SF $73) you can probably afford to build up a 10 year supply, especially with volume discounts.
Espressif is usually very good with both pricing and availability if you wish to buy in bulk.
I bought 1000 ESP32-S2 chips (not modules) for a project, it cost me a grand total of 1200 bucks as a nobody. During the height of the shortage from most western manufacturers too!
> If you build something that has to guarantee a 10 or 20 year supply
Consumer products routinely drop support after 3/5 years, Android phones being exhibit A, Sonos speakers are exhibit B, and Google in general being in the corner of shame.
While I applaud any efforts at product longevity, I am not getting the impression that it's a first-priority concern at present.
Sonos dropped support for speakers that were 10+ years old (and even then, still kept around the old app to mostly support them). Not sure what you’re on about with a 3 year lifespan…
That's neither here nor there. I have speakers that are 16 years old, Creative Gigaworks T3. In their 6th year, the circuit board in their sound remote died, rendering them useless. There are no complex chips there, nothing proprietary, but Creative no longer manufactured it, 5 years old, and so repair was impossible.
I was able to fix them by finding someone who sells a replacement remote on TaoBao and getting a shipping company to send it to me at 2x the price.
Point is - availability of parts does not guarantee manufacturer has any interest is supporting the product.
On the contrary, if manufacturer wants to fix something, they actually can, even if exact same parts aren't available.
Yes, I've used it in a commercial product with 10000+ deployments. It was the only chip with BLE and WiFi, so there was no other option at the time. If the requirements were different, I would use something from Nordic Semiconductors [0] or some ARMv8 chip.
The hardware itself is fine, but the biggest pain was getting stable WiFi and BLE connections simultaneously, because of only one antenna/radio. RAM was also a problem, it would be great to have at least 512kb. The SDK from Espressif is sometimes a little bit weird, but usable and bugs are fixed quickly. The build system is ok, nothing special.
The original ESP32 also has 512KB of RAM but some is reserved for instruction and flash cache. So on both the OG and the S3 you end up with about 350KB usable. Wifi is manageable but bluetooth will use most of it.
That said both the original ESP32 and the S3 support external PSRAM (4MB* for the og, 32MB for the S3).
You can buy modules with such RAM already (which you would likely do for most deployment), but if you really need a single chip solution they also sell the ESP32 chips with that "external" RAM embedded in it (albeit in smaller qty, like 2MB).
* The OG actually supports 8MB but only the first 4MB is directly addressable, you have to handle bankswitching yourself to access more.
Oh I think I got confused with the ESP32s with integrated "external" ram. And yeah, I think I was comparing them to the esp8xxx, which obviously have less ram. Still, the esp32 hasn't changed that much since the first version then I guess. I mean except for all the peripherals!
We started an ESP32-based design but for internal R&D purposes, within our inkjet manufacturing division. In that sense it's not commercial but uses professional toolsets and contract manufacturing etc., assurance of supply and supply chains.
In the past we mainly used STM32 both because it's a great chip and because we have a tight relationship with ST. However, some of the younger employees we recently brought on are big fans of the ESP32 and we thought "why not give the youngsters a chance".
It's looking really promising, the ESP32 nicely fills the niche between low-level Arduino-like micros and Linux-based larger boards like the Pi.
I’ve designed a few boards with ESP32s on them. Sometimes as a secondary system to provide wifi/ble, sometimes as the main microcontroller. One example was a commercial display using these P4 panels to display an 8-bit themed animation, and another one where a bunch of old small green CRTs were wired together to display the same “command prompt”-style text displaying the opening hours and other information.
STM32 would be the other big family of microcontrollers. And if you can get your hands on them, and need the vastly more powerful capabilities, the i.MX series can be interesting.
Not yet (in development), but there are literally 10's of millions of these powering common commercialized devices at this point.
I have several controllable LED light bulbs that are all ESP-based, bought on Amazon. Lot's of sensors, and general business and consumer oriented stuff using ESP8266 and ESP32 chips.
My dad has a little gadget for his big green egg that monitors and adjusts the temperature by controlling a small fan at the air intake. You can check the temperature over WiFi either through their app or on a very basic webpage the device serves. I peeked inside once and it's definitely a basic ESP8266 module mounted to a board. It's the perfect module for this kind of application. An ESP32 would be a little bit of overkill for this kind of application unless you wanted to add a display or some other extra processing load to the hardware. A module from Nordic would also be appropriate.
I'm not sure why this keeps coming up. Why wouldn't ESP32 be used in commercial products??
>Also, what would be considered the go-to chip for commercial applications?
ESP32, if it fits the product needs.
There are as many types of microcontrollers as there are needs for specific functionality. ESP32 fits practically any application that needs a microcontroller and wifi or bluetooth. That's the description of quite a lot of IoT products.
Cheap ICs and hobbyists-only do not go together. Cheap hobbyists ICs are almost invariably cheap because there's substantial commercial volume, and someone figured they could work for hobbyists as well.
AFAIK Shelly is the biggest customer of ESP32 chips. They sell cheap but relatively reliable smart home WiFi devices with a local web server that are really popular.
I think it is more like a C3 with Wifi6 than a C6. It has the same clocked RISCV processor and lacks the C6 low power processor and Zigbee baseband/mac. It has more peripherals than C3 but fewer than C6. It has an interestingly named "Power Glitch Detector" unlike the C6's and C3's "Brownout Detector."
They could just choose to always leave the DFS 5 GHz frequencies disabled but I don't think that's really a huge burden one way or the other after building the rest of a Wi-Fi chip and stack.
While I’m glad there progressing I suspect I’ll continue to buy 2.4ghz wifi 5 ones. I park all my IOT stuff on that to keep 5 and 6 ghz clean for key device.
How does the shared single core impact the application vs having a dedicated core on the original ESP32?
Also, why did they move to a single core, as I've seen a die shot of the ESP32, and the cores were only a fraction. 90% of the die went to RAM and the radio.
Other ESP32's in the range don't get more than about 20Mbps throughput under ideal conditions, so it probably isn't exactly a beast in the wifi performance department.
In fact, I wouldn't be surprised if this isn't the same silicon with an upgraded firmware to enable this new functionality.