Questions on direct-cabling 3x3 MIMO systems (ath9k)

Questions on direct-cabling 3x3 MIMO systems (ath9k)

[Ben Greear]

I finally got my 3-attenuator system up and running,
and I'm starting to do some tests.

AP and Station are running 3.5.5+, ath9k ar9380 NICs.
Channel 157, HT40, no encryption.

One question right away:  Should I expect decent performance
if I directly cable 2 wifi NICs, where one is acting as AP and
the other as station?  I'm cabling ch0 to ch0, ch1 to ch1, ch2 to ch2.

If I add about 15dBm of attenuation in this setup, max download
speed is about 70Mbps, though the station may train up to 450Mbps
rate.  Signal is reported as -23, and noise at -93.

Throughput stays about the same all the way to signal of -68 or so, though
the trained rate drops to around 270Mbps.

I can't make the signal quality go lower (even by attaching terminators
directly to the antenna mounts on the AP).  Guess I'll need some sort of
box to double-box these systems and hopefully drown out more of the RF.

Thanks,
Ben

-- 
Ben Greear <greearb@candelatech.com>
Candela Technologies Inc  http://www.candelatech.com

Re: Questions on direct-cabling 3x3 MIMO systems (ath9k)

[Ben Greear]

On 10/04/2012 11:49 AM, Ben Greear wrote:
> I finally got my 3-attenuator system up and running,
> and I'm starting to do some tests.
>
> AP and Station are running 3.5.5+, ath9k ar9380 NICs.
> Channel 157, HT40, no encryption.
>
> One question right away:  Should I expect decent performance
> if I directly cable 2 wifi NICs, where one is acting as AP and
> the other as station?  I'm cabling ch0 to ch0, ch1 to ch1, ch2 to ch2.

Well, to try to answer my own question:

I mainly needed to tune socket tx/rx-buffer sizes, and some similar
things to get the performance up to expected rates, so it was
mostly user error.

Now, I'm still seeing lots of strange issues when running through
the attenuator with lots of attenuation, but at least when things
are configured for low amounts of attenuation, I get good performance
(330+Mbps UDP throughput).

Off to look at excessive crc errors...

Thanks,
Ben

-- 
Ben Greear <greearb@candelatech.com>
Candela Technologies Inc  http://www.candelatech.com

Re: Questions on direct-cabling 3x3 MIMO systems (ath9k)

[Julian Calaby]

Hi Ben,

On Fri, Oct 5, 2012 at 4:49 AM, Ben Greear <greearb@candelatech.com> wrote:
> I finally got my 3-attenuator system up and running,
> and I'm starting to do some tests.
>
> AP and Station are running 3.5.5+, ath9k ar9380 NICs.
> Channel 157, HT40, no encryption.
>
> One question right away:  Should I expect decent performance
> if I directly cable 2 wifi NICs, where one is acting as AP and
> the other as station?  I'm cabling ch0 to ch0, ch1 to ch1, ch2 to ch2.

Going off on a tangent: arguably it shouldn't matter which channel on
the AP card is connected to which channel on the STA card. I wonder
what results you'd get if you cabled them in each combination and
tested the throughput keeping the other variables the same. I also
wonder what impact it would have on the throughput if you started
disconnecting the cables.

Thanks,

-- 
Julian Calaby

Email: julian.calaby@gmail.com
Profile: http://www.google.com/profiles/julian.calaby/
.Plan: http://sites.google.com/site/juliancalaby/

Re: Questions on direct-cabling 3x3 MIMO systems (ath9k)

[Ben Greear]

On 10/04/2012 03:36 PM, Julian Calaby wrote:
> Hi Ben,
>
> On Fri, Oct 5, 2012 at 4:49 AM, Ben Greear <greearb@candelatech.com> wrote:
>> I finally got my 3-attenuator system up and running,
>> and I'm starting to do some tests.
>>
>> AP and Station are running 3.5.5+, ath9k ar9380 NICs.
>> Channel 157, HT40, no encryption.
>>
>> One question right away:  Should I expect decent performance
>> if I directly cable 2 wifi NICs, where one is acting as AP and
>> the other as station?  I'm cabling ch0 to ch0, ch1 to ch1, ch2 to ch2.
>
> Going off on a tangent: arguably it shouldn't matter which channel on
> the AP card is connected to which channel on the STA card. I wonder
> what results you'd get if you cabled them in each combination and
> tested the throughput keeping the other variables the same. I also
> wonder what impact it would have on the throughput if you started
> disconnecting the cables.

I doubt it matters either....just seemed sane to start with something
I could easily keep track of in my head :)

I had really shitty performance when I had only one cable connected
to the station, but I think it is probably related to rate-control,
which seems too aggressive.  When I force things to slower speeds
it works fine.  I'm starting to look into that now.

I hope to eventually produce big pretty graphs reporting signal, rx-rate,
etc over various attenuations...but still got a ways to go first!

Thanks,
Ben

-- 
Ben Greear <greearb@candelatech.com>
Candela Technologies Inc  http://www.candelatech.com

Re: Questions on direct-cabling 3x3 MIMO systems (ath9k)

[Julian Calaby]

Hi Ben,

On Fri, Oct 5, 2012 at 8:41 AM, Ben Greear <greearb@candelatech.com> wrote:
> On 10/04/2012 03:36 PM, Julian Calaby wrote:
>>
>> Hi Ben,
>>
>> On Fri, Oct 5, 2012 at 4:49 AM, Ben Greear <greearb@candelatech.com>
>> wrote:
>>>
>>> I finally got my 3-attenuator system up and running,
>>> and I'm starting to do some tests.
>>>
>>> AP and Station are running 3.5.5+, ath9k ar9380 NICs.
>>> Channel 157, HT40, no encryption.
>>>
>>> One question right away:  Should I expect decent performance
>>> if I directly cable 2 wifi NICs, where one is acting as AP and
>>> the other as station?  I'm cabling ch0 to ch0, ch1 to ch1, ch2 to ch2.
>>
>>
>> Going off on a tangent: arguably it shouldn't matter which channel on
>> the AP card is connected to which channel on the STA card. I wonder
>> what results you'd get if you cabled them in each combination and
>> tested the throughput keeping the other variables the same. I also
>> wonder what impact it would have on the throughput if you started
>> disconnecting the cables.
>
>
> I doubt it matters either....just seemed sane to start with something
> I could easily keep track of in my head :)

Fair enough.

> I had really shitty performance when I had only one cable connected
> to the station, but I think it is probably related to rate-control,
> which seems too aggressive.  When I force things to slower speeds
> it works fine.  I'm starting to look into that now.
>
> I hope to eventually produce big pretty graphs reporting signal, rx-rate,
> etc over various attenuations...but still got a ways to go first!

Pretty graphs are always useful. It'd be nice to compare other brands
/ models of card too, but I'm guessing that's outside the scope of
what you're planning to achieve.

Are you planning to open-source any of the software / hardware you're
making to do this?

Thanks,

-- 
Julian Calaby

Email: julian.calaby@gmail.com
Profile: http://www.google.com/profiles/julian.calaby/
.Plan: http://sites.google.com/site/juliancalaby/

Re: Questions on direct-cabling 3x3 MIMO systems (ath9k)

[Ben Greear]

On 10/04/2012 03:53 PM, Julian Calaby wrote:
> Hi Ben,
>
> On Fri, Oct 5, 2012 at 8:41 AM, Ben Greear <greearb@candelatech.com> wrote:
>> On 10/04/2012 03:36 PM, Julian Calaby wrote:
>>>
>>> Hi Ben,
>>>
>>> On Fri, Oct 5, 2012 at 4:49 AM, Ben Greear <greearb@candelatech.com>
>>> wrote:
>>>>
>>>> I finally got my 3-attenuator system up and running,
>>>> and I'm starting to do some tests.
>>>>
>>>> AP and Station are running 3.5.5+, ath9k ar9380 NICs.
>>>> Channel 157, HT40, no encryption.
>>>>
>>>> One question right away:  Should I expect decent performance
>>>> if I directly cable 2 wifi NICs, where one is acting as AP and
>>>> the other as station?  I'm cabling ch0 to ch0, ch1 to ch1, ch2 to ch2.
>>>
>>>
>>> Going off on a tangent: arguably it shouldn't matter which channel on
>>> the AP card is connected to which channel on the STA card. I wonder
>>> what results you'd get if you cabled them in each combination and
>>> tested the throughput keeping the other variables the same. I also
>>> wonder what impact it would have on the throughput if you started
>>> disconnecting the cables.
>>
>>
>> I doubt it matters either....just seemed sane to start with something
>> I could easily keep track of in my head :)
>
> Fair enough.
>
>> I had really shitty performance when I had only one cable connected
>> to the station, but I think it is probably related to rate-control,
>> which seems too aggressive.  When I force things to slower speeds
>> it works fine.  I'm starting to look into that now.
>>
>> I hope to eventually produce big pretty graphs reporting signal, rx-rate,
>> etc over various attenuations...but still got a ways to go first!
>
> Pretty graphs are always useful. It'd be nice to compare other brands
> / models of card too, but I'm guessing that's outside the scope of
> what you're planning to achieve.
>
> Are you planning to open-source any of the software / hardware you're
> making to do this?

I'm going to publish the code for the attenuator (including
eagle layout & gerber files for the Arduino shield).  Will market them
for sale as completed units too.  Will publish a simple command-line
tool to adjust the attenuation, and unit will have knobs to turn
so you don't really need a computer to control it anyway.

The user-space code that handles traffic generation and
reporting will remain proprietary, but someone that didn't
mind playing with iperf or similar should be able to reproduce
at least the simpler test cases that I plan to do fairly easily.

We'll be able to test various APs..but on the client side,
I doubt we'll test other than ath9k anytime soon, although
I think our software would support other WiFi NICs if you didn't
try to do any of the virtual interface stuff.

Thanks,
Ben


-- 
Ben Greear <greearb@candelatech.com>
Candela Technologies Inc  http://www.candelatech.com

Re: Questions on direct-cabling 3x3 MIMO systems (ath9k)

[Julian Calaby]

Hi Ben,

On Fri, Oct 5, 2012 at 9:01 AM, Ben Greear <greearb@candelatech.com> wrote:
> On 10/04/2012 03:53 PM, Julian Calaby wrote:
>>
>> Hi Ben,
>>
>> On Fri, Oct 5, 2012 at 8:41 AM, Ben Greear <greearb@candelatech.com>
>> wrote:
>>>
>>> On 10/04/2012 03:36 PM, Julian Calaby wrote:
>>>>
>>>>
>>>> Hi Ben,
>>>>
>>>> On Fri, Oct 5, 2012 at 4:49 AM, Ben Greear <greearb@candelatech.com>
>>>> wrote:
>>>>>
>>>>>
>>>>> I finally got my 3-attenuator system up and running,
>>>>> and I'm starting to do some tests.
>>>>>
>>>>> AP and Station are running 3.5.5+, ath9k ar9380 NICs.
>>>>> Channel 157, HT40, no encryption.
>>>>>
>>>>> One question right away:  Should I expect decent performance
>>>>> if I directly cable 2 wifi NICs, where one is acting as AP and
>>>>> the other as station?  I'm cabling ch0 to ch0, ch1 to ch1, ch2 to ch2.
>>>>
>>>>
>>>>
>>>> Going off on a tangent: arguably it shouldn't matter which channel on
>>>> the AP card is connected to which channel on the STA card. I wonder
>>>> what results you'd get if you cabled them in each combination and
>>>> tested the throughput keeping the other variables the same. I also
>>>> wonder what impact it would have on the throughput if you started
>>>> disconnecting the cables.
>>>
>>>
>>>
>>> I doubt it matters either....just seemed sane to start with something
>>> I could easily keep track of in my head :)
>>
>>
>> Fair enough.
>>
>>> I had really shitty performance when I had only one cable connected
>>> to the station, but I think it is probably related to rate-control,
>>> which seems too aggressive.  When I force things to slower speeds
>>> it works fine.  I'm starting to look into that now.
>>>
>>> I hope to eventually produce big pretty graphs reporting signal, rx-rate,
>>> etc over various attenuations...but still got a ways to go first!
>>
>>
>> Pretty graphs are always useful. It'd be nice to compare other brands
>> / models of card too, but I'm guessing that's outside the scope of
>> what you're planning to achieve.
>>
>> Are you planning to open-source any of the software / hardware you're
>> making to do this?
>
>
> I'm going to publish the code for the attenuator (including
> eagle layout & gerber files for the Arduino shield).  Will market them
> for sale as completed units too.  Will publish a simple command-line
> tool to adjust the attenuation, and unit will have knobs to turn
> so you don't really need a computer to control it anyway.

Nice! I wish I had a good reason to build one =)

> We'll be able to test various APs..but on the client side,
> I doubt we'll test other than ath9k anytime soon, although
> I think our software would support other WiFi NICs if you didn't
> try to do any of the virtual interface stuff.

Makes sense.

Thanks,

-- 
Julian Calaby

Email: julian.calaby@gmail.com
Profile: http://www.google.com/profiles/julian.calaby/
.Plan: http://sites.google.com/site/juliancalaby/

Re: Questions on direct-cabling 3x3 MIMO systems (ath9k)

[Ben Greear]

On 10/04/2012 04:48 PM, Julian Calaby wrote:
> Hi Ben,
>

>>> Pretty graphs are always useful. It'd be nice to compare other brands
>>> / models of card too, but I'm guessing that's outside the scope of
>>> what you're planning to achieve.
>>>
>>> Are you planning to open-source any of the software / hardware you're
>>> making to do this?
>>
>>
>> I'm going to publish the code for the attenuator (including
>> eagle layout & gerber files for the Arduino shield).  Will market them
>> for sale as completed units too.  Will publish a simple command-line
>> tool to adjust the attenuation, and unit will have knobs to turn
>> so you don't really need a computer to control it anyway.
>
> Nice! I wish I had a good reason to build one =)
>
>> We'll be able to test various APs..but on the client side,
>> I doubt we'll test other than ath9k anytime soon, although
>> I think our software would support other WiFi NICs if you didn't
>> try to do any of the virtual interface stuff.
>
> Makes sense.

Ok, here's a link that has some pics of the attenuator and a graph
showing link speed, throughput rate, and rx-signal quality.  I have
lots more testing of various combinations, but plan to spend more time
on automating it before I do much more manual testing.

I'll not spam the list further with such links, but feel free to check
back on my blog...I plan to post more results there as I get the time.

http://bens-workshop.blogspot.com/2012/10/programmable-attenuator-arduino-shields.html

If anyone else has any similar data, I'd love to see it posted somewhere.

Thanks,
Ben


-- 
Ben Greear <greearb@candelatech.com>
Candela Technologies Inc  http://www.candelatech.com