Performance v/s signal quality graph.

[Ben Greear]

On 08/05/2014 08:15 AM, Bart Jooris wrote:
> Hi Ben,
> 
> I enjoyed watching your graph already a few times now ;-)
> 
> I have a few questions related to your experiment:
> -about the setup, the cards have 3x3 MIMO so the attenuator has probably also three channels which are attenuated in parallel?

Yes.  It is possible for us to attenuate the individual channels differently, but we
have not tried doing any of that type of testing in our automated reports.

> -based on the datasheet of the QCA9880 VHT80_MCS9 reception should break around -68dBm. On your graph it breaks around -52dBm, which is not within the described
> tolerances. Do you have any idea what could be causing this? Is the noise level higher than expected? Or did I made the wrong conclusions? I will try to verify
> on our setup.

Since it shows such good throughput at very hot signal, maybe the signal reporting is off by
a bit, shifted lower, perhaps?  I don't know how to make ath10k report the noise (haven't looked
to figure out if it even can with existing firmware.)

> -my million dollar question to you is related to the approx. 810Mbps UDP goodput :-) I already did some experiments over 3 coax cables, I noticed the same UDP
> goodput as over the air, which is around 640Mbps. Watching the sniffer I've noticed MCS=9, Data rate=1299.9 Mbps and block ACKs every (in avg.) 10 frames. So
> what is the key to the higher goodput?

For us, the key is using a 3.7Ghz E5 Intel processor :)

We do not see such good throughput on a mobile Core i7 processor system,
for instance.

I was just trouble-shooting one of our systems at a customer's site and they are getting up to 870Mbps UDP goodput
through their AP (seems to range from around 800 to 870 in their case...I do not know their physical layout,
but probably it is cabled up.)

> -related to the hot signal, without having proof I've noticed that  it is easier to have higher goodputs with a low tx-power set, like 1 dBm. But probably more
> related to our setup... Which tx-power did you use during your experiments? Would you be able to get the same results (point break around -52dBm) with another
> tx-power set?

I have not tried setting different power levels, so it is using defaults.  I'll make a note to run some additional
tests at different power levels.  I think that it should not matter much since we are putting a variable attenuator
inline, but with wifi, thinking doesn't always match reality I've found!

Thanks,
Ben

> 
> Thanks,
> 
> Bart
> 
> 
> On 07/26/2014 01:15 AM, Ben Greear wrote:
>> Some time back, I ran some tests, and someone (sorry, can't remember or find the email)
>> suggested my signal might be too hot.  I was using a 30 db fixed attenuator inline with the SMA
>> cables.
>>
>> I set up a test today with E5 processor system as station, and a fast i7 processor
>> system as AP, with the two systems cabled through our variable attenuator.  Open authentication,
>> channel 149, other wifi equipment in area on 5ghz was (mostly) idle or powered down.
>> Kernel is 3.14.13+, latest ath10k CT (10.1.467 based) firmware on both systems
>> (though I don't think that matters much for throughput tests).  10 virtual stations
>> configured on station machine, but only one was actively doing traffic.
>>
>> Here is a udp download v/s rx-signal graph as the attenuator is stepped from 0 to 95.5
>> db attenuation in .5 db steps.
>>
>> http://www.candelatech.com/pdfs/ath10k_atten_vs_throughput.pdf
>>
>> The rx-signal axis is the left hand side.  The green graph is the station's (ie netdev)
>> rx bps, the black is UDP goodput.
>>
>> It seems ath10k (WLE900VX) has no real trouble with hot signals, and runs best between
>> -15 and -52 rx signal level (as reported by the station device).
>>
>> Our systems are not perfectly isolated, so the maximum attenuation is around -85.
>>
>> Enjoy,
>> Ben
>>
> 
> 


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