Build your own WAP like Ubiquity WAP

Hardware Questions and Recommendations
1

Dear community,

I have been looking on the internet to find an answer why it seems to be so hard to build your own WAP ? Everyone just says buy a WAP since it is build for it and then install OpenWrt on top of it.
However, I was wondering why it is so hard to build one yourself ? In my non-expert WAP brain I don't see why we can build a proper OPNsense router on a mini-pc but cannot build a proper OpenWrt access point on top of a "mini-WAP-pc" with some WAP-wifi card ?

Please help me understand why it seems so difficult to build your own WAP ? I mean all the other hundreds of companies have also done it, why can't the average Joe buy some WAP parts and build one themselves ?

Please don't answer with: Why would you want to build it yourself ? It is so cheap to buy a premade WAP!

The reason I ask is because I want to learn from it.
Thanks in advance guys!!!

Cheers!!! ;);):wink:

1 Like
2

Sure you can. You can dig up the issues people encounter while trying that here on the forum - recent enough wireless cards that don't draw too much power (or SBCs that do provide enough power), driver issues, people thinking Intel client cards work in AP mode, ...

It's a fun timesink/learning experience if you're up for it. Chances that you will obtain something with a smaller footprint and lower power consumption and lower BOM than off the shelf hardware are close to zero though.

Companies have experience in board design and power management, access to manufacturer documentation and whatnot.

3 Likes
3

Thank you for your reply @Borromini,

What I don't understand is why is there no information available on how to make something like that in the open ? it is 2025 hahah, I mean there must have been someone from a WAP company that shared this knowledge on how to build this right ? I did not find it but well hahaha. Why are there no board designs WAPs available, documentation, power management available ? Why is a mini-pc for a router easier to build than a good working WAP ?

I mean the guys that build ubiquity WAPs for instance also needed to learn it from somewhere :wink:

thanks again!

4

I'd argue much the same for wired routers built on commodity hardware - especially when using x86. It's just easier because regular ethernet does not have the concept of AP and client.

1 Like
5

Are you talking about assembling off the shelf hardware, or building a high performance AP?

It's not hard to cobble together a bunch of parts into a case, but you'll be hard pressed to get a really high performance system. Why?? Read on...

To build a proper AP, that requires actual electrical, and critically RF design expertise, not to mention the other aspects of building hardware such as enclosure design with proper thermal management and a myriad of other things.

But let's focus on RF design, since that's the primary purpose of an AP. Designing very high performance RF circuits and antenna systems is the domain of PhD's and engineers with considerable industry experience. While a lot of product specific design might be confidential company information and/or trade secret, the general knowledge isn't kept secretive insofar as anybody can buy text books and/or enroll in classes about RF and antenna design. But, you need to have graduate level understanding of the underlying mathematics and core electrical engineering design principles to understand and apply the principles.

Once you're done learning, you need to have the software and equipment to design and measure the RF circuitry and antenna systems. To be fully equipped, you're likely talking about a $100K USD (likely upwards of $500K to $1M USD) investment. The software is expensive, the test equipment (including Faraday cages and even RF anechoic chambers) is even more so.

Then you need to understand the regulatory landscape of the countries in which you plan to sell your product. You'll need to design around that and then perform pre-certification testing until you have your design designed dialed in. Finally, you submit your product for official tests for each region. This process adds significant costs to your product design process, but also requires a lot of knowledge to navigate these regulations (both technical and legal).

Finally, you can manufacture your product at scale... that means spinning up a factory line. And, for RF, that also means that you have designed test and calibration routines that can run on a per-unit basis on the factory floor such that you can guarantee both regulatory compliance and high performance.

I didn't even discuss the low level firmware development and so many other factors, but even this list should hopefully show that RF devices are quite complicated to build at a commercial level.

Now... all that said, this has been a discussion about making a very high performance, highly integrated device like you see from the likes of UI, Cisco, Aruba, TP-Link, Netgear, etc...

It's not to say that you cannot build a home-brew solution with off-the-shelf parts that is reasonably performant.... you could even design your own PCB and use some off-the-shelf antennas. Yes... it's possible to make an AP at home... but you'll be pretty hard pressed to get the high end performance that come with arrays of custom antennas that are fully calibrated to get the most out of multiple spatial streams and beamforming capabilities of the commercially designed and manufactured integrated systems.

Does that make sense?

7 Likes
6

While everything psherman has raised is correct, I'll expand a bit on this.

A client centric WLAN M.2 PCIe card (e.g. mt7921) costs around ~25 EUR +/- (yes, there is cheaper, and more expensive if you look towards 6 GHz or wifi7, but this is a rule of thumb). They come as 2x2, often with a passive heatsink (so heat dissipation is already a topic) and can do either 2.4 GHz XOR 5 GHz (XOR 6 GHz for mt7921k or better). While these can be used for AP mode, they are not made with that in mind and come with additional limitations (max number of connected clients, range, etc.).

AP centric cards (e.g. mt7915, QCN5024/ QCN5054/ QCN9074) are a different beast

  • physically larger (often beyond M.2 sizes, you sometimes still see mini-PCIe here)
  • 4x4
  • much higher power consumption (3.3V/ 3A ~= 10 watts and more, sometimes with aux power)
  • accordingly higher heat dissipation, passive heatsink required - airflow necessary (good thermal design --> convection or active cooling)
  • usually not available locally, you generally do have to import them yourself (the market for these is too small)
  • pricing around >60 EUR plus shipping (mt7915), up to 200-250 EUR plus shipping (qcn5024/ qcn5054/ qcn9074)
  • you will need to buy pigtails (4* per card) and antennas separately (4* per card), quality varies a lot (especially for 5/ 6 GHz), so expect to buy from multiple sources to get the best ones
  • you do depend on the vendor having done their homework and programmed the EFUSEs (wireless features and restrictions) and EEPROM (wireless calibration data) correctly, or the card won't work properly (this a recurring probem, so you need to buy from reputable sources, who will 'fix it' (~replace the card))
  • multiply this by two (2.4+5 GHz) or three (2.4+5+6 GHz), after all if you go this route, you surely do want 4x4 and concurrent tri-band and not replicate the feature set of a 15 buck AP.

Since 802.11ac, (Mu-)MIMO and beamforming are relevant, so you do need proper spatial distribution of your antennas (correct distance, polarization), see the black magic psherman laid out (high frequency antenna design is a higher art). Attaching these to the PCIe brackets of your ATX mainboard will not give you great results, they're:

  • too close to each other for beamforming/ Mu-MIMO to work
  • create interference with themselves and the other bands
  • the huge metal case they're attached to (PC case) isn't exactly great either (shielding a ~150°-210° angle from good reception)
  • as mentioned, power delivery (beyond spec) and heat dissipation is a problem

So from an antenna design point of view, you do want a 'small' case, preferably with little metal - but at the same time, you're hard pressed to find small x86_64 mainboards (<< µATX) with enough := three PCIe/ M.2 slots for your three WLAN cards. Again, heat dissipation is a problem, these cards will cook themselves to death/ crash if there isn't enough airflow, as is power delivery - the cards alone will eat up 30 watts (on top of the power consumption of the rest of your system, so ~50-60 watts idle at best, more if you want to dabble into 10 GBe as well), even in idle mode (there is no real idle in AP mode).

Can you do it, sure - if you have deep pockets.
Does it make sense if you're a wireless developer, yes (but in that case a single good wireless card might be sufficient, tri-band -but not concurrent tri-band), then you don't have much of a choice anyways (it does help massively if you can compile/ bisect locally on the host) - and maybe a client-centric mt7621k card already suffices here (you're experimenting, not serving 30+ stations with your AP).

So if you want to go this route, you need:

  • at least a µATX board with >= 3 PCIe slots (respectively a SFF system with a proprietary board), mITX and smaller boards with sufficient PCIe ports exist, but are really expensive (Xeon D and similar), look at >=600 EUR
  • hope/ test that it can deliver enough power (may go up in smoke, literally)
  • buy 3* AP mode centric WLAN cards (concurrent 2.4+5+6 GHz), at 60-250 EUR a pop (so you've spent between 180-650 EUR on wireless cards alone)
  • buy 3*4 pigtails, for around a fiver each (~60 EUR) - quality of the feedlines is important, especially for >=5 GHz, so don't skimp on prices here, be prepared that your first choice is $NOT_GOOD and that you have to get a better set
  • buy 3*4 antennas, for around a fiver each (~60 EUR) - quality of the antennas is important, especially for >=5 GHz, so don't skimp on prices here, be prepared that your first choice is $NOT_GOOD and that you have to get a better set
  • experiment -a lot- to get decent reception results with the placement of your twelve antennas

Ignoring the cost of the actual PC here, you are looking at an investment -for the wireless cards, pigtails, antennas only- for somewhere between 300-770 EUR here, ignoring shipping costs, customs fun and having to replace some of your purchases becase they may not be good enough. As well as an idle power consumption of >50-60 watts, 24/7 (which adds up as well). Sure, you can do that - or you can buy e.g. an Acer Predator W6 (AXE7800) and flash OpenWrt for ~180 EUR delivered.

  • yes, these tri-band plastic routers are expensive as well (300-400 EUR is easily possible)
  • yes, they do eat electricity for breakfast as well (~20-30 watts idle is common)
  • yes, you can only choose between less than a handful of supported alternatives (with varying regional availability)
  • yes, some of them come with active cooling/ fans
  • but they are 'small' (at least in comparison) and 'neat' (at least if you're into spaceship design or upside down spiders…)
  • and you can expect that the vendor has done a decent hf design, they went through the certification process and didn't just stick the antennas where the PCIe bracket told them to
    meaning you can expect Mu-MIMO and beamforming to work
  • and you can expect it to work, even under OpenWrt (for supported devices)
    • afaik there are still problems if you want to operate multiple QCA/ ath11k PCIe cards in one system, so another thing you'd have to sort out yourself for your self-build system

For an AP, the wireless performance matters most -CPU performance only to a much lesser extent- small and neat are advantages (being able to place it when it makes most sense or even screwing it to a wall). It is hard and expensive to beat even a 20 buck mt7621a+mt7615DBDC device in this regard (and if you get multiple of those for your home, to place in the various storeys or into the wings of your castle, you will easily beat any single highest-end AP in practice).

It's very hard to compete with a mass produced AP in terms of cost and actual performance - and if you want to try, you really do need to have deeeep pockets, lost of time and patience for experimenting to get best results and love talking to companies in Singapore or China and your local customs agents. Can it be done, also with OpenWrt, yes.

Times have changed since 802.11n and before, back then the distinction between client- and AP centric WLAN cards didn't exist (APs were equipped with bog standard PCMCIA or mini-PCI cards, literally), you only had to deal with 2.4 GHz (if you added 5 GHz to the mix, you already faced similar issues as the ones described above, albeit to a much lesser degree, simpler technology) and maybe half a dozen of clients to cater for.

Nowadays, beyond the 'simple' WLAN world, you also have to deal with interference from bluetooth, zigbee, z-wave, proprietary wireless streaming technologies (headphones, wireless HDMI transceivers, proprietary wireless surveillance cameras) and even in small homes you may easily see >>25 WLAN devices on the air at all time (more than triple that for smarthome enthusiasts). And be aware that wifi 7 is already peaking at us, with wifi 8 being already entering press coverage. If you are an enthuasiast (and if you want to build your own AP, you are), you might get 4-5 years of service from your self-built system, before you have to start all over again with the next iteration of the standards.

Looking back, 802.11b, 802.11g and 802.11n only brought you small improvements in practice (also because early draft-n devices were bad™, buggy and didn't achieve good performance). 802.11ac and newer are a real game changer.

  • good wifi 4 may give you ~120 MBit/s (2x2 client, 5 GHz, HT40) in practice - maybe 60-70 MBit/s over 2.4 GHz
  • wifi 5 will give you 250-350 MBit/s (2x2 client, 5 GHz, VHT80)
  • wifi 6 will give you 700-800 MBit/s (2x2 client, 5 GHz, HE80)
  • wifi 7 can give you well over 2 GBit/s in practice (2x2 client, 5+6 GHz MLO, EHT160)
6 Likes
7

Deep analysis has been shown above, to give you a general idea on this is, wireless hardware is a combination of card + antennae design, there are tons of different brands/models of router/AP on market but in fact the underlying wireless cards are with only a limited set, why do one model is having better range than others (e.g. GLINET MT3000 vs Cudy WR3000, later one has much better coverage while they use the same wireless board). Most people don't see the antennae design part but in fact it's one of the most important factor, which cannot be done easily by layman.

You can ignore form factor, and/or power of the device, but if you consider signal coverage as important factor, you will find it difficult to achieve by yourself.

8

That is a very important aspect as well - and looks can be deceiving as well. Many large antennas on the outside do not necessarily beat good internal antennas (my strongest devices has 16 internal antennas, distributed at various angles) - and one of my stronger devices even has flimsy internal PCB antennas. HF antenna design is black magic, illogical and requires a lot of studying and institutional knowledge, as well as lots of experimenting in a shielded hf chamber to get it right. You can only recoup these r&d costs, if you sell batches of tens of thousands of your devices.

9

Wauwww!! literally wauw!! for the detailed information you guys have provided!! really sweet of you!!! @psherman @slh @fakemanhk

I never heard about RF engineering in my life and did not know it was a field on its own. Really interesting. While dissecting your guys input I did stumble upon some motherboards, wifi cards and antennas that were fairly cheap and that seem to work pretty good.

For instance this company: https://www.wallystech.com/

Sells AP motherboards, wifi cards and antennas. I have seen some videos on youtube that actually install OpenWRT on these devices and they seem to work pretty good. I guess what I meant with building these devices yourself is if the motherboard, wifi card and antenna's are already made. But to completely start from scratch that would be quite something.

Cheers!!!

10

When I looked at my Buffalo WXR-5950AX12 vs Netgear WAX220 vs Netgear WAX206, I really feel it. The later 2 are designed to be "AP" and Netgear really did a good job on it.

Even I'm a graduate of Electronics Engineering, with certain knowledge in this field, I don't want to do it myself. Yes you did quote companies that sell antennae, but antenna that is "just working" vs "best performance" are totally different things. Sometimes you might heard about "high gain antenna", "unidirectional", etc....they are specially designed for different purposes, on market many sellers are just selling you things that are "just working". If your plan is to create AP for a small coverage without many clients, usually this is enough, so it's up to you to decide.

3 Likes
11

I don't really understand how you get from "build your own AP" to buying a devboard clone, putting it into some kind of crude casing and sticking random antennas onto it.

That's pretty much the worst combination (unless you need a devboard, different use case, very different), as it's more expensive, there's little for you to decide (so none of 'build your own') and all the downsides of unmatched antenna design, as well as basic OpenWrt support at best (it's a devboard, not a product).

1 Like
12

He wants to build a custom router, OpenWrt One but with replaceable/upgradeable hardware.

Think of Homeassistant Yellow as one such example, you can upgrade the Raspberry Pi CM on it.

Similarly, you could upgrade from a Filgoic 820 to an 830 processor just by replacing a daughter board. change the radio to Wifi 7be from a Wifi 6 radio etc.

13

This is where the problem coming from, did you read the above comments about RF engineering? Do you think 5GHz and 6GHz can use the same antenna? Obviously not, will it work? Yes but the performance won't be optimal.

Homeassistant example is irrelevant, since it has no radio part, you should probably think of changing "Bluetooth" to "Zigbee"....something like that.

1 Like
14

Bluetooth to Zigbee is easy these days.

I am personally using the Dual Bluetooth/Zigbee firmware on the same thermometer. it is detected by both integrations. this is not something I will use in the production system but at least it exists and works.

When changing the radios we will need to change the antennas at the same time for more than just working performance.

what we are discussing will be enthusiast devices not meant for business use but for hardware hackers.

15

Bluetooth/Zibee are still not exactly the same frequency band (Zigbee has more coverage), what I mean here is unlike most cabled systems, wireless can be very different between generations, it's not easy to design an optimal one, of course just like what I mentioned above, if you want something "just work", then go ahead, but don't come back and ask "why the coverage isn't great".

16

that is exactly my point, this hackable hardware isn't meant for production/business use and this should be pointed out very clearly on the box, experimental hardware.
Range and WIfi speed are not guaranteed.

it could be antennas + radio on the same replaceable radio modules.

17

But unless you massively overspec the mainboard and pay through through the nose, you can't do that (at least not with good results).

In wifi 5 times, mini-PCIe was still common, no one knew that wifi 6 cards would be M.2 and chug 10 watts from the connector. And if you want to choose and mix components, you're also in for quite some surprises on the software side as well (device tree, pinmux, hardware quirks and plain -bugs, these are embedded devices, not a PC (and for the PC side, see above)). None of the boards mentioned by the OP (of that nasty spamming company) would qualify for that.

The BPi-r4 might, for Mediatek wifi 6-7 cards, maaaybe wifi 8 (and that's going to be very questionable), but at that point it will be a low-end mainboard.

Just look at the Turris Omnia, high-end when it came out, expensive, wifi 5. The company did retrofit it to wifi 6, but that was not easy (and many cards didn't work), the effort was plagued by mini-PCIe slots, cooling/ power-consumption of the wifi 6 cards, now-exotic mSATA storage, and the SOC is still ARMv7/ 32 bit, so upstream software support is fading (kvm, etc.).

At some point it may be better to just buy a plastic enclosure, 50*50*50 cm³, mount a mains power plug inside and place the current mid-range consumer router inside - replace it with the then-current mid-range model in 3-4 years and pretend that you've 'upgraded' it. Schrödinger's router in a box - cheaper and better/ more consistent performance over the whole time.

2 Likes
18

Lack of information and support.

Lower footprint is absolutely possible and the reason I myself went the DIY route. Sometimes, lower footprint is much more important than price...

However, in my current build I cheaped out on WLAN card - by getting a DBDC MT7915 for ~$37.5 as a stopgap until 802.11be is well supported with reasonably priced AP cards available. Been waiting over a year for that to happen, still no solution...

19

I remember when I saw the $37.5 7915DBDC card, I went with the Netgear WAX206 (which also has MT7915, plus 2.5GbE x 1) and it was roughly $30, later I helped a friend to buy Cudy WR3000 which has MT7976CN built-in for something like $50. Recently got the Netgear WAX220 PoE ceiling mount AP which is no more than $60, I don't need to worry about the signal coverage since all WiFi design are there already.

If you look at posts here, asking about using USB WiFi, or other self-built thing, they really meant for production (at least production in their home), and there could be lots of follow up questions asking "why this is not working", well people tend to forget that "you're being warned"

1 Like
20

You're probably from the US. Where I live, WAX206 cost $150+ when it was available - and WAX220 never was. At least Cudy WR3000 is ~$55. Regardless, all the devices mentioned dwarf my build, which also has a WWAN.

I bought MT7915 with couple months to test the waters in mind before switching to 802.11be, though - not 1.5+ years of everyday use. Didn't expect to use OpenWRT either, (naively) thought I'd just setup everything under a clean general purpose distribution & lock it down hard...