WARP Project Forums - Wireless Open-Access Research Platform

The Rx PHY calculates its channel estimates from the long training symbols in the preamble. The LTS is defined as BSPK values in the frequency domain on 52 subcarriers (48 data, 4 pilots). The channel estimator computes the average of the two estimates per subcarrier (one per LTS copy). The resulting per-subcarrier complex scalar is used by the equalizer to demodulate each subcarrier. All 52 complex values are also stored with the packet for retrieval by the MAC code.

The magnitude of the complex scalar channel estimate corresponds to the magnitude of that subcarrier's channel coefficient. You can safely compare the magnitudes of estimates for different subcarriers in one packet to establish relative SNRs between subcarriers.

To compute an absolute SNR you must account for the gains applied by the Rx amplifiers. The Rx PHY operates on whatever samples are output from the ADC. The PHY does not care what the gain settings were[1], only that the digital samples approximately fill the ADC's dynamic range. Thus, to compute an absolute power for a signal in the PHY you must include the gains that were applied to the signal before it reached the Rx PHY. The LNA gains are approximately [0, 15 30]dB; the VGA gains are approximate [0, 63]dB. The MAX2829 datasheet has actual plots of voltage gain vs. gain setting for these amps.

[1] Not quite true- the gain selections are used to select the higher-SNR antenna when selection diversity is enabled.

Does this mean that we can only get the relative SNR between different subcarriers, but not the (average or total?) SNR across all subcarriers for the entire packet?

The sum the magnitudes of all 52 channel estimates will correspond to the magnitude of the analog signals at the ADCs. If you then adjust for the Rx gains, you would get an estimate for the amplitude of the signal at the RF front end, integrated over the frequencies of the 52 subcarriers (i.e. ±8 MHz around the center frequency). This value will correlate highly with the RSSI-based Rx power estimate.

Under which scenarios does the former relative SNR metric become useful? Can we use it, for example, to do adaptive sub-carrier bit loading?

Yeah, exactly. Unequal modulation or power across subcarriers could exploit differences in magnitudes of subcarrier channel coefficients. Actually building this is hard, since it requires feedback of per-subcarrier channel state information within the coherence time of the channel.

The best way to get an SNR estimate is probably to add logic to calculate Error Vector Magnitude for each symbol. See this discussion for more details.