You are not logged in.

**shanggdlk****Member**- Registered: 2017-Apr-12
- Posts: 22

Hi, I'm using WARP V3 board and WARPLab package.

I run the SISO ofdm example and plot the phase reading of the computed channel state ( %Calculate channel estimate from average of 2 training symbols

350 rx_H_est = lts_f .* (rx_lts1_f + rx_lts2_f)/2;). While the figure shows that the phase reading of the 54 subcarriers changes across the whole 2*PI three times. However, in theory, the phase reading of the 52 subcarriers in 20M bandwidth will not change so much (just around a PI). I know the delay from Tx->Rx and the instantaneous phase offset due to CFO also contributes a little bit. But I did not experiment in a quite small room (3mx3m), so I believe the delay may also not contribute too much to the phase change. Could you tell me why the phase across the whole 20M band changes so much?

Another question is in the Long Training Symbol, why not use all 64 subcarriers for channel estimation?

Thanks!

Offline

**murphpo****Administrator**- From: Mango Communications
- Registered: 2006-Jul-03
- Posts: 5159

One possibility is the FFT_OFFSET parameter. This value sets how many samples of the cyclic prefix are used in the FFT of each OFDM symbol. In theory you can use 0 samples of the cyclic prefix. However because the LTS correlation can be wrong by 1-2 sample periods, it's better to use a few samples of cyclic prefix for synchronization tolerance. Using 1 sample of cyclic prefix trades 1 sample period of delay spread tolerance for 1 sample of synchronization error. In the vast majority of propagation channels, this is a safe tradeoff.

Using cyclic prefix samples in the Rx FFT is equivalent to a delay from Tx -> Rx and will manifest as a linearly-increasing phase across subcarriers, where the slope of the phase is proportional to the "delay" (i.e. number of cyclic prefix samples). This phase offset is normally estimated as part of the overall channel response and is equalized by the Rx processing. As long as the channel delay spread is less than the unused cyclic prefix, this scheme incurs no penalty.

Offline

**shanggdlk****Member**- Registered: 2017-Apr-12
- Posts: 22

THanks, Murphpo.

By the way, why not use 64 subcarriers for channel estimation? I try to modify the LTS, but it seems the phrase reading of the subcarrier [1,18-28] does not change linearly as other subcarriers change. Is it any other way to get relatively accurate channel estimation across the whole 64 subcarriers? thanks!

Offline

**shanggdlk****Member**- Registered: 2017-Apr-12
- Posts: 22

## murphpo wrote:

One possibility is the FFT_OFFSET parameter. This value sets how many samples of the cyclic prefix are used in the FFT of each OFDM symbol. In theory you can use 0 samples of the cyclic prefix. However because the LTS correlation can be wrong by 1-2 sample periods, it's better to use a few samples of cyclic prefix for synchronization tolerance. Using 1 sample of cyclic prefix trades 1 sample period of delay spread tolerance for 1 sample of synchronization error. In the vast majority of propagation channels, this is a safe tradeoff.

Using cyclic prefix samples in the Rx FFT is equivalent to a delay from Tx -> Rx and will manifest as a linearly-increasing phase across subcarriers, where the slope of the phase is proportional to the "delay" (i.e. number of cyclic prefix samples). This phase offset is normally estimated as part of the overall channel response and is equalized by the Rx processing. As long as the channel delay spread is less than the unused cyclic prefix, this scheme incurs no penalty.

THanks, Murphpo.

By the way, why not use 64 subcarriers for channel estimation? I try to modify the LTS, but it seems the phrase reading of the subcarrier [1,18-28] does not change linearly as other subcarriers change. Is it any other way to get relatively accurate channel estimation across the whole 64 subcarriers? thanks!

Offline

**murphpo****Administrator**- From: Mango Communications
- Registered: 2006-Jul-03
- Posts: 5159

By the way, why not use 64 subcarriers for channel estimation? I try to modify the LTS, but it seems the phrase reading of the subcarrier [1,18-28] does not change linearly as other subcarriers change. Is it any other way to get relatively accurate channel estimation across the whole 64 subcarriers? thanks!

The WARPLab OFDM examples use the same subcarrier map as the 802.11a/g waveform. This spec reserves the upper/lower subcarriers to meet the spectral mask requirement for 20MHz channels. You can experiment with non-zero values in these subcarriers by modifying the LTS definition to have +1 or -1 values. It is critical the LTS have good auto-correlation properties in the time domain, so be sure to compare your custom LTS to the standard LTS at the output of the correlation step.

Offline