looking for suggestions for a small router/appliance board/SoC

[Jim Thompson]

> On Oct 22, 2015, at 2:39 PM, John Nielsen <lists at jnielsen.net> wrote:
> 
> On Oct 22, 2015, at 12:19 PM, Jim Thompson <jim at netgate.com> wrote:
> 
>>> On Oct 22, 2015, at 12:57 PM, John Nielsen <lists at jnielsen.net> wrote:
>>> 
>>> Hi-
>>> 
>>> I’m working on a proof-of-concept for a kind of networking swiss army knife. Can anyone suggest a board that meets the following requirements? CPU arch doesn’t matter as long as it will run FreeBSD (Atom, ARM, MIPS, etc).
>>> 
>>> - Small form factor (SoC, probably)
>>> - Can support at least 2 802.11a/b/g/n adapters, prefer 3 (any combination of chip-integrated and mini PCI-e slots. Prefer to avoid USB if possible)
>> 
>> I openly question your need or the desirability for 3 802.11 adapters.  It can be made to work, but you’re going to have some intermod.
> 
> I don’t mind being questioned. :) I haven’t yet had to worry much about intermod; can you educate me? One or two of the radios would be in the 5GHz band at any given time. One scenario (out of several) where I envisioned having 3 radios is taking a wireless uplink (STA in either 2.4 or 5 GHz band) and repeating it (HOSTAP) on both 2.4 and 5 GHz. Totally crazy?

Many people believe that there are “three non-overlapping” channels for use with 802.11b/802.11g/802.11n in 2.4GHz.

While the transmit masks don’t overlap, the selectivity of the receiver (especially after the industry turned to direct conversion architectures around
the advent of 802.11g) is not sufficient to operate even two radios in any given bad (2.4GHz, etc.)

It’s all been covered before.  You’ve unknowingly hit “Jim’s favorite point of banter”.

https://lists.freebsd.org/pipermail/freebsd-wireless/2013-December/004158.html
http://seclists.org/interesting-people/2009/Oct/77
http://www.ietf.org/mail-archive/web/manet/current/msg05757.html
http://lists.shmoo.com/pipermail/hostap/2004-April/006524.html

Even where you can operate in separate bands, there are mixing products that can greatly interfere with correct reception.

Receivers live under constant bombardment of signals which enter through the antenna port. Some of these signals are immediately attenuated due to front-end filtering, (aka pre-selection). When the remaining signals reach a non-linear element, such as a detector, mixer or amplifier, harmonics of the signals are generated. Most of the harmonics are well outside the pass band of RF and IF filters and cause no problems.

However there are some frequencies where the mixing products – intermodulation products – of the various signals fall on or near the desired receive frequency range. The intermodulation products that tend to cause the most problems are the so-called odd-order products. This is true because odd-order products of signals near your desired receive frequency also are near your receive frequency. 802.11 systems tend to suffer more from these issues due to the uniform spacing of the channels.  (Remember that OFDM separates the signal out into many sub-channels.)

With DC receivers, the intermediate frequency is zero and the image to the desired channel (for all but single-sideband signals) is the channel itself. This means 
only one local oscillator (LO) is required, which means only one phase noise contribution, and as such, the need for the bulky off-chip filters is consequently removed.  Filtering now only occurs at low frequencies (baseband) with some amplification, which means less current consumption than at higher frequencies (to drive device parasitics), fewer components and lower cost.  This is all good, and contributes the much lower cost for today’s 802.11 radios, though the largest contributor here is the absence of SAW filters at the IF in a superhet receiver.

Practically, however, strong out-of-band interference or blocking signals may need to be removed prior to down-conversion in order to avoid desensitizing the receiver by saturating subsequent stages, as well as producing harmonics and intermodulation terms which will then appear in the baseband.  

In direct conversion, as the signal of interest is converted to baseband very early in the receive chain, without any filtering other than RF band-selection, various phenomena contribute to the creation of DC signals, which directly appear as interfering signals in the band of interest.  The LO may be conducted or radiated through an unintended path to the mixer's RF input port, thus effectively mixing with itself, producing an unwanted DC component at the mixer output. Worse still, this LO leakage may reach the LNA input, producing an even stronger result. This effect presents a high barrier against the integration of LO, mixer and LNA on a single silicon substrate, where numerous mechanisms can contribute to poor isolation. These include substrate coupling, ground bounce, bond wire radiation, and capacitive and magnetic coupling.  I’ve seen LO signals cross over the PCI bus lines.  (Vivato had a design with 14 802.11b (superhet) radios on the same PCB.  I improved that to 6 802.11g (DC, Atheros) radios on discrete cards.)

Conversely, a strong in-band interference signal, once amplified by the LNA, may find a path to the LO-input port of the mixer, thus once again producing self-mixing. 

Some amount of LO power will be conducted through the mixer and LNA (due to their non-ideal reverse isolation) to the antenna. The radiated power, appearing as an interferer to other receivers in the corresponding band, may violate emissions standards of the given system. It is important to note that since the LO frequency is inside the receive band, the front-end filters do nothing to suppress this LO emission. Additionally, the radiated LO signal can then be reflected by buildings or moving objects and re-captured by the antenna. This effect, however, is not of significant importance compared to the aforementioned LO self-mixing and blocking signal self-mixing. 

The leakage of LO or RF signals to the opposite mixer port is not the only way in which unwanted DC can be produced. Any stage that exhibits even-order nonlinearity will also generate a DC output.

Suffice it to state that many people have this idea, but few actually endure the engineering to make it a) work and b) be legal in a given (set of) regulatory domain(s).


>>> - Has or supports at least 2 1GbE ports. Prefer 3-5 ports with switching functionality
>>> - Storage not super constrained. Built-in storage (if any) can be small (which I’m arbitrarily defining as less than 128MB) if there is also an SD card slot or similar. USB storage will do in a pinch.
>>> - Has at least 2 free USB ports after meeting previous requirements
>>> - Serial port or header (or GPIO pins that can be used as one? Not too familiar with that)
>>> - Low power consumption (within reason taking the above into account)
>>> - Low cost (again, within reason)
>>> 
>>> I may just start with a PC Engines apu1d, but if there are boards that are smaller, cheaper, have lower power requirements and/or have integrated wifi or switch capabilities I’d like to look in to them as well.
>> 
>> you can probably get APU in low-volume, but quantity is very constrained these days (it’s been true all year).
> 
> Good to know.
> 
>> We do have the RCC-VE and RCC-DFF units available.
>> 
>> http://store.netgate.com/Desktop-Systems-C83.aspx
> 
> Thanks for the link!

You could also look at Minnowboard Max (quite difficult to get) or Minnowboard Turbot.
http://www.minnowboard.org/meet-minnowboard-max/