Re: Is the data rate estimation for 5GHz channels overly pessimistic?

[James Prestwood <prestwoj@gmail.com>]

Hi Simonas,

On 10/16/23 4:35 AM, James Prestwood wrote:
> Hi Somonas,
> 
> On 10/15/23 12:40 PM, Simonas Kazlauskas wrote:
>> Hey,
>>
>> Denis Kenzior wrote:
>>> It may be we screwed something up with HE since hardware was still 
>>> pretty rare when the feature was developed.  But we do treat RSSI 
>>> values below -82 as almost unusable.  Perhaps this needs to be 
>>> tweaked for newer hardware.
>>>
>>> If you want to get your hands dirty, it should be fairly easy to 
>>> modify unit/test-band.c to test what our estimate code does with your 
>>> specific circumstances.
>>>
>>> The local HE capabilities can be found via 'iw phy', looking at what 
>>> iwd prints at start, or using iwmon.  Similarly, remote capabilities 
>>> can be sniffed using iwmon and running an iwd scan or 'iw scan trigger'.
>>>
>>> Me or James can walk you through all this if needed.
>>
>> Thanks for offering. I have cracked open the file, and just started 
>> playing around by adding a new `he_test_data` structure. Alas, I’m 
>> really hazy on what I ought to be filing into the `capabilities` and 
>> `he_capabilities` fields. My initial guess is something like this, 
>> based on one of the other tests around:
>>
>> +/* 5GHz, 40MHz, MCS 11, NSS 1 */
>> +const struct he_test_data he_test_5_40mhz_mcs_5_nss_1 = {
>> +    .freq = BAND_FREQ_5_GHZ,
>> +    .rssi = -76,
>> +    .expected_rate = 275200000ULL,
>> +    .capabilities = {
>> +        .he_mcs_set = { HE_MCS_SET(MCS11, 1), MCS_UNSUP },
>> +        .he_phy_capa = { HE_PHY_CAPA(0x00) },
>> +        .iftypes = 1 << NETDEV_IFTYPE_STATION,
>> +    },
>> +    .he_capabilities = {
>> +        22, IE_TYPE_HE_CAPABILITIES - 256, HE_MAC_CAPA,
>> +        HE_PHY_CAPA(0x00), MCS_UNSUP, HE_MCS_SET(MCS11, 1)
>> +    },
>> +};
>>
>> This computes a rate of `25800000`, which I’m guessing corresponds to 
>> 25.8Mbps, perhaps? That’s fair I guess, since the only thing I really 
>> really changed in there compared to the other tests is the `MCS11` 
>> (that’s supported by AX201) and the `rssi`. There clearly are plenty 
>> of other fields that I should be filling in here, I feel, if I wanted 
>> to accurately represent my hardware.
>>
>> I think I managed to grab the capabilities of both the AP and the 
>> client using your instructions, but I’m really unclear on how to go 
>> from the output from `iwmon` to fields in this struct. My best guess 
>> right now is to actually run `iwd` under a gdb and place some 
>> breakpoints in the scanning code to extract the capabilities. Would 
>> that make sense, or is there a better way to fill in these fields?
>>
>> (As I’m writing this mail, I realize now that I probably want to 
>> somehow specify NSS=2 for `2x2` MIMO…?)
>>
>> Another thing that confuses me a great deal is the fact that the scan 
>> results with my APs don’t actually show any of the HE capabilities, 
>> only the HT and VHT ones (see attached sample from the `iwmon` 
>> output.) I also tried scanning an Android hotspot on a reasonably 
>> recent phone, and although the set of MCS supported were different, 
>> the output still contains only HT and VHT capabilities. I would love 
>> to attach my pcap here, but unfortunately I also have no clue how to 
>> sanitize it, so I’m going to attach an excerpt from `iwmon`’s output 
>> for the time being. Is this expected, or could this perhaps be at 
>> least part of the problem I’m seeing? Is it possible that an AP could 
>> not advertise HE capabilities at scan time, but for HE to still be 
>> used after authentication occurs?
> 
> Looks like iwmon doesn't actually print out the HE element, but that 
> doesn't explain the same behavior on the android device. If your more 
> comfortable you could send the pcap just to Me or Denis.
> 
> But as Denis pointed out maybe this needs another look as far as signal 
> strength any usability. The difficult part is one card could perform 
> much better at -85dbm than another. I'll take a closer look and see if 
> something got missed in the spec and do some testing myself (I think I 
> have an ax201, and definitely an ax210).

FWIW, my 802.11ax mesh APs I have at home don't advertise the HE element 
either.

> 
> Thanks,
> James
> 
>>
>> Thank you,
>> S.