[Make-wifi-fast] Wiring up a wireless testbed

[Make-wifi-fast] Wiring up a wireless testbed

[Toke Høiland-Jørgensen]

Since I had to (physically) move my wireless testbed recently, I had to
figure out a way to run reliable WiFi experiments in a cramped server
room. I ended up wiring everything up instead of running over the air,
and documented the process here, in case anyone wants to replicate it:

https://blog.tohojo.dk/2017/11/building-a-wireless-testbed-with-wires.html

Also, if anyone sees any fatal flaw in that setup, please do let me know :)

-Toke

Re: [Make-wifi-fast] Wiring up a wireless testbed

[Aaron Wood]

[-- Attachment #1: Type: text/plain, Size: 1197 bytes --]

What this setup also removes is random noise, which is good for
repeatability, but bad for real-life testing. My experiences doing similar
are that there isn’t much of a grey area between a “great”
throughput/packet error rate and “nothing works”.

It also removes the real-world crosstalk between the antennas.

So good for some things, especially automated tests, but it’s not a
replacement for real field tests in the presence of random noise.
On Fri, Nov 3, 2017 at 15:51 Toke Høiland-Jørgensen <toke@toke.dk> wrote:

> Since I had to (physically) move my wireless testbed recently, I had to
> figure out a way to run reliable WiFi experiments in a cramped server
> room. I ended up wiring everything up instead of running over the air,
> and documented the process here, in case anyone wants to replicate it:
>
> https://blog.tohojo.dk/2017/11/building-a-wireless-testbed-with-wires.html
>
> Also, if anyone sees any fatal flaw in that setup, please do let me know :)
>
> -Toke
> _______________________________________________
> Make-wifi-fast mailing list
> Make-wifi-fast@lists.bufferbloat.net
> https://lists.bufferbloat.net/listinfo/make-wifi-fast

[-- Attachment #2: Type: text/html, Size: 1735 bytes --]

Re: [Make-wifi-fast] Wiring up a wireless testbed

[Bob McMahon]

[-- Attachment #1: Type: text/plain, Size: 2347 bytes --]

A way I think about is to emulate the "system" in both range/power
(attenuation) and in mixing (phase shifts)   It's a diagonal matrix for
range feeding into an h-matrix for the mixing.  A butler matrix can be used
for the latter if variable phase control isn't required, e.g. you don't
care about spatial stream relative powers.

On adding energy for "random" noise, some knobs of concern are the energy
detect on the tx and signal floor for the rx.   These sources can be fed
into the same h-matrix through their own d-matrices.  This won't replace
field tests but helps get closer towards that while providing for repeat
ability.

Also, equally important by my judgment, though not related to wireless, is
to synchronize the clocks on the PCs.  An oven controlled oscillator and
PTP works well towards that goal.

Bob



On Fri, Nov 3, 2017 at 4:03 PM, Aaron Wood <woody77@gmail.com> wrote:

> What this setup also removes is random noise, which is good for
> repeatability, but bad for real-life testing. My experiences doing similar
> are that there isn’t much of a grey area between a “great”
> throughput/packet error rate and “nothing works”.
>
> It also removes the real-world crosstalk between the antennas.
>
> So good for some things, especially automated tests, but it’s not a
> replacement for real field tests in the presence of random noise.
>
> On Fri, Nov 3, 2017 at 15:51 Toke Høiland-Jørgensen <toke@toke.dk> wrote:
>
>> Since I had to (physically) move my wireless testbed recently, I had to
>> figure out a way to run reliable WiFi experiments in a cramped server
>> room. I ended up wiring everything up instead of running over the air,
>> and documented the process here, in case anyone wants to replicate it:
>>
>> https://blog.tohojo.dk/2017/11/building-a-wireless-
>> testbed-with-wires.html
>>
>> Also, if anyone sees any fatal flaw in that setup, please do let me know
>> :)
>>
>> -Toke
>> _______________________________________________
>> Make-wifi-fast mailing list
>> Make-wifi-fast@lists.bufferbloat.net
>> https://lists.bufferbloat.net/listinfo/make-wifi-fast
>
>
> _______________________________________________
> Make-wifi-fast mailing list
> Make-wifi-fast@lists.bufferbloat.net
> https://lists.bufferbloat.net/listinfo/make-wifi-fast
>

[-- Attachment #2: Type: text/html, Size: 3445 bytes --]

Re: [Make-wifi-fast] Wiring up a wireless testbed

[Dane Medic]

[-- Attachment #1: Type: text/plain, Size: 795 bytes --]

What value do you use for TX power in this antenna-less setup?

On 3 November 2017 at 23:51, Toke Høiland-Jørgensen <toke@toke.dk> wrote:

> Since I had to (physically) move my wireless testbed recently, I had to
> figure out a way to run reliable WiFi experiments in a cramped server
> room. I ended up wiring everything up instead of running over the air,
> and documented the process here, in case anyone wants to replicate it:
>
> https://blog.tohojo.dk/2017/11/building-a-wireless-testbed-with-wires.html
>
> Also, if anyone sees any fatal flaw in that setup, please do let me know :)
>
> -Toke
> _______________________________________________
> Make-wifi-fast mailing list
> Make-wifi-fast@lists.bufferbloat.net
> https://lists.bufferbloat.net/listinfo/make-wifi-fast

[-- Attachment #2: Type: text/html, Size: 1409 bytes --]

Re: [Make-wifi-fast] Wiring up a wireless testbed

[Toke Høiland-Jørgensen]

Aaron Wood <woody77@gmail.com> writes:

> What this setup also removes is random noise, which is good for
> repeatability, but bad for real-life testing. My experiences doing similar
> are that there isn’t much of a grey area between a “great”
> throughput/packet error rate and “nothing works”.
>
> It also removes the real-world crosstalk between the antennas.
>
> So good for some things, especially automated tests, but it’s not a
> replacement for real field tests in the presence of random noise.

Yeah, I agree that real-world testing is needed on top of this. But for
my own needs I consider repeatability more important than strict
correspondence to a real-world scenario (whatever that means). But
you're right, of course, that it would be bad if this is the only type
of testing that is done :)

As for cross-talk between antennas, I'm not sure this setup removes it
completely; there's probably leakage around connectors, and possibly in
the cables attaching the outwards-facing SMA connectors on each machine
to whatever the tiny connectors on the cards are called. Don't have the
equipment to measure any of that, though, so for now it remains a hunch...

-Toke

Re: [Make-wifi-fast] Wiring up a wireless testbed

[Toke Høiland-Jørgensen]

Bob McMahon <bob.mcmahon@broadcom.com> writes:

> A way I think about is to emulate the "system" in both range/power
> (attenuation) and in mixing (phase shifts)   It's a diagonal matrix for
> range feeding into an h-matrix for the mixing.  A butler matrix can be used
> for the latter if variable phase control isn't required, e.g. you don't
> care about spatial stream relative powers.
>
> On adding energy for "random" noise, some knobs of concern are the energy
> detect on the tx and signal floor for the rx.   These sources can be fed
> into the same h-matrix through their own d-matrices.  This won't replace
> field tests but helps get closer towards that while providing for repeat
> ability.

What equipment would it take to do something like that?

> Also, equally important by my judgment, though not related to wireless, is
> to synchronize the clocks on the PCs.  An oven controlled oscillator and
> PTP works well towards that goal.

Yeah, already using PTP in my testbed; quite essential for measuring
one-way delay.

-Toke

Re: [Make-wifi-fast] Wiring up a wireless testbed

[Toke Høiland-Jørgensen]

Dane Medic <dm70dm@gmail.com> writes:

> What value do you use for TX power in this antenna-less setup?

I haven't touched the TX power settings, so it's doing whatever the
driver does by default. Which I think is 30dBm at the frequencies I'm
using (5Ghz, channel 165).

-Toke

Re: [Make-wifi-fast] Wiring up a wireless testbed

[Aaron Wood]

[-- Attachment #1: Type: text/plain, Size: 727 bytes --]

> As for cross-talk between antennas, I'm not sure this setup removes it
> completely; there's probably leakage around connectors, and possibly in
> the cables attaching the outwards-facing SMA connectors on each machine
> to whatever the tiny connectors on the cards are called. Don't have the
> equipment to measure any of that, though, so for now it remains a hunch...
>

Yeah, that's going to depend on the noise-floor of the receivers.  I've
found that -105dBm receive radios (other technologies) needed physical
separation and cables rated at 60dB of shielding in order to keep
cross-talk down.

The cables from the SMA connectors to the boards are likely pretty leaky
(the board connectors are probably u.fl connectors)

[-- Attachment #2: Type: text/html, Size: 1015 bytes --]

Re: [Make-wifi-fast] Wiring up a wireless testbed

[Toke Høiland-Jørgensen]

Aaron Wood <woody77@gmail.com> writes:

>> As for cross-talk between antennas, I'm not sure this setup removes it
>> completely; there's probably leakage around connectors, and possibly in
>> the cables attaching the outwards-facing SMA connectors on each machine
>> to whatever the tiny connectors on the cards are called. Don't have the
>> equipment to measure any of that, though, so for now it remains a hunch...
>>
>
> Yeah, that's going to depend on the noise-floor of the receivers.  I've
> found that -105dBm receive radios (other technologies) needed physical
> separation and cables rated at 60dB of shielding in order to keep
> cross-talk down.

Yeah, I've heard similar reports from others. And I'm not really in a
position to set that up; nor do I really need that much isolation
between the nodes (I think).

> The cables from the SMA connectors to the boards are likely pretty
> leaky (the board connectors are probably u.fl connectors)

Ah, yes, u.fl is what those are called. Indeed they are. So yeah, those
cables don't exactly look like they have a lot of shielding on them... :)

-Toke

Re: [Make-wifi-fast] Wiring up a wireless testbed

[Bob McMahon]

[-- Attachment #1: Type: text/plain, Size: 1757 bytes --]

Try telemakus <http://www.telemakus.com/> for variable attenuators and
variable phase shifters.    Aeroflex works well to though are a bit more
expensive, e.g. the 8331
<http://weinschel.apitech.com/weinschel/pdfiles/wmod8331.pdf> is an example

Power dividers, splitter/combiners and butler matrix devices are sold by
multiple vendors.  Prices will vary. Unfortunately, prices aren't super
low.  But usually it's worth the extra money to get a reliable and
controllable test bed, particularly when one values their time into the
equation.

Bob

On Sat, Nov 4, 2017 at 2:48 PM, Toke Høiland-Jørgensen <toke@toke.dk> wrote:

> Bob McMahon <bob.mcmahon@broadcom.com> writes:
>
> > A way I think about is to emulate the "system" in both range/power
> > (attenuation) and in mixing (phase shifts)   It's a diagonal matrix for
> > range feeding into an h-matrix for the mixing.  A butler matrix can be
> used
> > for the latter if variable phase control isn't required, e.g. you don't
> > care about spatial stream relative powers.
> >
> > On adding energy for "random" noise, some knobs of concern are the energy
> > detect on the tx and signal floor for the rx.   These sources can be fed
> > into the same h-matrix through their own d-matrices.  This won't replace
> > field tests but helps get closer towards that while providing for repeat
> > ability.
>
> What equipment would it take to do something like that?
>
> > Also, equally important by my judgment, though not related to wireless,
> is
> > to synchronize the clocks on the PCs.  An oven controlled oscillator and
> > PTP works well towards that goal.
>
> Yeah, already using PTP in my testbed; quite essential for measuring
> one-way delay.
>
> -Toke
>

[-- Attachment #2: Type: text/html, Size: 2359 bytes --]

Re: [Make-wifi-fast] Wiring up a wireless testbed

[Bob McMahon]

[-- Attachment #1: Type: text/plain, Size: 2318 bytes --]

Also, Ixia is selling a controllable RF system and it's in early stages.
Maybe contact them to see what the market is driving them to build?   My
guess is their asking price will exceed $100K and has huge margins relative
to bill of materials.

One could probably work with telemakus and build a few using their parts
and sell it for cheaper than the IXIA chassis and still make a nice profit.

Bob

On Sat, Nov 4, 2017 at 6:50 PM, Bob McMahon <bob.mcmahon@broadcom.com>
wrote:

> Try telemakus <http://www.telemakus.com/> for variable attenuators and
> variable phase shifters.    Aeroflex works well to though are a bit more
> expensive, e.g. the 8331
> <http://weinschel.apitech.com/weinschel/pdfiles/wmod8331.pdf> is an
> example
>
> Power dividers, splitter/combiners and butler matrix devices are sold by
> multiple vendors.  Prices will vary. Unfortunately, prices aren't super
> low.  But usually it's worth the extra money to get a reliable and
> controllable test bed, particularly when one values their time into the
> equation.
>
> Bob
>
> On Sat, Nov 4, 2017 at 2:48 PM, Toke Høiland-Jørgensen <toke@toke.dk>
> wrote:
>
>> Bob McMahon <bob.mcmahon@broadcom.com> writes:
>>
>> > A way I think about is to emulate the "system" in both range/power
>> > (attenuation) and in mixing (phase shifts)   It's a diagonal matrix for
>> > range feeding into an h-matrix for the mixing.  A butler matrix can be
>> used
>> > for the latter if variable phase control isn't required, e.g. you don't
>> > care about spatial stream relative powers.
>> >
>> > On adding energy for "random" noise, some knobs of concern are the
>> energy
>> > detect on the tx and signal floor for the rx.   These sources can be fed
>> > into the same h-matrix through their own d-matrices.  This won't replace
>> > field tests but helps get closer towards that while providing for repeat
>> > ability.
>>
>> What equipment would it take to do something like that?
>>
>> > Also, equally important by my judgment, though not related to wireless,
>> is
>> > to synchronize the clocks on the PCs.  An oven controlled oscillator and
>> > PTP works well towards that goal.
>>
>> Yeah, already using PTP in my testbed; quite essential for measuring
>> one-way delay.
>>
>> -Toke
>>
>
>

[-- Attachment #2: Type: text/html, Size: 3318 bytes --]

Re: [Make-wifi-fast] Wiring up a wireless testbed

[Bob McMahon]

[-- Attachment #1: Type: text/plain, Size: 2992 bytes --]

Also, the final piece of the puzzle is chip power consumption.    Monsoon
makes a current meter for a reasonable price.

I realize the focus here is on low latency.   The market is also driving
power consumption, i.e. "energy consumed for useful information energy
successfully delivered with the lowest latency possible."  The chips
themselves can give the RF energies (as well as dump h matrices) so having
an expensive network analyzer for that isn't required beyond chip design.

Bob

On Sat, Nov 4, 2017 at 6:56 PM, Bob McMahon <bob.mcmahon@broadcom.com>
wrote:

> Also, Ixia is selling a controllable RF system and it's in early stages.
> Maybe contact them to see what the market is driving them to build?   My
> guess is their asking price will exceed $100K and has huge margins relative
> to bill of materials.
>
> One could probably work with telemakus and build a few using their parts
> and sell it for cheaper than the IXIA chassis and still make a nice profit.
>
> Bob
>
> On Sat, Nov 4, 2017 at 6:50 PM, Bob McMahon <bob.mcmahon@broadcom.com>
> wrote:
>
>> Try telemakus <http://www.telemakus.com/> for variable attenuators and
>> variable phase shifters.    Aeroflex works well to though are a bit more
>> expensive, e.g. the 8331
>> <http://weinschel.apitech.com/weinschel/pdfiles/wmod8331.pdf> is an
>> example
>>
>> Power dividers, splitter/combiners and butler matrix devices are sold by
>> multiple vendors.  Prices will vary. Unfortunately, prices aren't super
>> low.  But usually it's worth the extra money to get a reliable and
>> controllable test bed, particularly when one values their time into the
>> equation.
>>
>> Bob
>>
>> On Sat, Nov 4, 2017 at 2:48 PM, Toke Høiland-Jørgensen <toke@toke.dk>
>> wrote:
>>
>>> Bob McMahon <bob.mcmahon@broadcom.com> writes:
>>>
>>> > A way I think about is to emulate the "system" in both range/power
>>> > (attenuation) and in mixing (phase shifts)   It's a diagonal matrix for
>>> > range feeding into an h-matrix for the mixing.  A butler matrix can be
>>> used
>>> > for the latter if variable phase control isn't required, e.g. you don't
>>> > care about spatial stream relative powers.
>>> >
>>> > On adding energy for "random" noise, some knobs of concern are the
>>> energy
>>> > detect on the tx and signal floor for the rx.   These sources can be
>>> fed
>>> > into the same h-matrix through their own d-matrices.  This won't
>>> replace
>>> > field tests but helps get closer towards that while providing for
>>> repeat
>>> > ability.
>>>
>>> What equipment would it take to do something like that?
>>>
>>> > Also, equally important by my judgment, though not related to
>>> wireless, is
>>> > to synchronize the clocks on the PCs.  An oven controlled oscillator
>>> and
>>> > PTP works well towards that goal.
>>>
>>> Yeah, already using PTP in my testbed; quite essential for measuring
>>> one-way delay.
>>>
>>> -Toke
>>>
>>
>>
>

[-- Attachment #2: Type: text/html, Size: 4409 bytes --]

Re: [Make-wifi-fast] Wiring up a wireless testbed

[Pete Heist]

> Date: Fri, 03 Nov 2017 23:51:05 +0100
> From: Toke Høiland-Jørgensen <toke@toke.dk>
> To: make-wifi-fast@lists.bufferbloat.net
> Subject: [Make-wifi-fast] Wiring up a wireless testbed
> Message-ID: <87tvybm0x2.fsf@toke.dk>
> Content-Type: text/plain
> 
> Since I had to (physically) move my wireless testbed recently, I had to
> figure out a way to run reliable WiFi experiments in a cramped server
> room. I ended up wiring everything up instead of running over the air,
> and documented the process here, in case anyone wants to replicate it:
> 
> https://blog.tohojo.dk/2017/11/building-a-wireless-testbed-with-wires.html
> 
> Also, if anyone sees any fatal flaw in that setup, please do let me know :)

That looks really nice, especially for reproducing and fixing software bugs. I’d second Aaron’s comments that what you get here is repeatability at the expense of not seeing things you see in the real world, which you’re probably aware of.

As an example of what I’ve seen, in the 2.4 GHz band with my OM2P-HS routers running LEDE, I sometimes see CTS protection suddenly flipping on or off in the middle of tests, changing throughput and results considerably. It doesn’t help that one distant neighbor seems to be using 40 MHz channels, so I’m sometimes struggling to find clear 2.4 GHz spectrum, but this is what can happen in the real world anyway.

Unfortunately, I don’t think I can replicate this “wired wireless" setup in my test closet at the moment, because none of my six APs (NanoStation M5s or OM2P-HSs) have external antenna connectors. Nor do the older MacBooks I started using for clients in point-to-multipoint testing. I’m not sure I’ll even do much more point-to-multipoint testing because of how disastrously the results vary between runs, or if I do, I sense it’s going to consume a lot of time. :)