WARP Project Forums - Wireless Open-Access Research Platform

Our current theory is that the AGC and/or packet detector stay locked when the RX-PHY is reset. This is validated by the observation that all packets are received if we set the pktDetector threshold high enough using

I don't think this is the cause. Toggling the Rx PHY reset via the WLAN_RX_REG_CTRL_RESET register will de-assert the OFDM packet detection latch, thereby clearing the pkt det state and resetting the AGC.

There is a post-reset timeout during which new packet detection events are blocked (this is to avoid pkt det events on Rx signal transients caused by the Rx amplifiers resetting to the default high gain state). After this timeout the pkt det logic is ready to re-assert on the next pkt det event. Also keep in mind that each pkt det event lasts at least the duration of the standard preamble. The Rx PHY will auto-reset the pkt det state if the LTS correlator does not assert at the expected time. Together these factors can lead to missed receptions of your desired waveform if it happens to overlap the first part of an interfering waveform (anytime during the preamble, SIGNAL field, SIGNAL decoding time or post-pkt-det-reset timeout).

There is a non-zero chance of the pkt det logic asserting in the middle of a reception, if the incoming waveform happens to have auto-correlation properties similar to the preamble STS. With the default thresholds this should be rare.

Some Wi-Fi chipsets implement "message in message", which attempts to detect the arrival of a stronger waveform during a weaker one and resets its Rx PHY to attempt decoding the stronger waveform. Our PHY does not implement this.

...all packets are received if we set the pktDetector threshold high enough using
wlan_mac_low_set_pkt_det_min_power();
wlan_phy_enable_req_both_pkt_det();

This makes sense. Enabling the RSSI detector and/or requiring a min power will reduce the PHY's sensitivity and can effectively filter out lower-power waveforms before the Rx PHY pipeline. If your interfering nodes are further away, you can configure the WARP node's Rx PHY to be "deaf" to the interference.

wlan_phy_rx_pktDet_autoCorr_ofdm_cfg(corr_thresh, energy_thresh, min_dur, post_wait)

in what unit is the energy_thresh parameter? Is there any way to directly correlate this with received packet power?

The unit is arbitrary; the energy thresh is compared to sum(I^2 + Q^2) in the auto-correlation pkt det pipeline. It needs to be small to allow detection of low-SNR waveforms, where the Rx I/Q don't fill the ADC range even with the AGC's default high gains.

Edlmann wrote:

In a second experiment, we moved to a free channel, and generated the long interference packet with a 3rd WARP Node.

...

would have expected a higher SINR to lead to better performance of our system.

What happens if you don't move to a different channel, but simply unplug the Wi-Fi equipment and replace it with the 3rd interfering WARP node? Could one explanation be that there are other sources of interference that you aren't accounting for? The tables you listed account for known signal powers and known interference powers (from the Wi-Fi link you control). But unknown interference powers could exist that wouldn't be tabulated there (e.g. a high power 802.11ac network, a microwave oven, etc.). Maybe the actual SINR isn't better in the WARP experiment but rather your experiment is blind to the other sources of interference.

If your Wi-Fi interfering equipment has external SMA antennas, you could control for external interference by switching away from an over-the-air experiment in favor of an over-the-wire experiment with RF cabling, attenuation, and a power splitter:

                 Interferer
                    |
                    A
                    |
TDMA1 ----A---- Power Splitter ----A----  TDMA2


A: Attenuator

The "interferer" here could be either the commercial Wi-Fi equipment or the WARPv3 interferer. This would let you repeat the same experiment but be much more confident about the source of the interference on the TDMA link.

I agree your observations suggest some difference in SINR vs PER between interference from your WARP vs commercial Wi-Fi equipment. But Chris' point is valid- confirming this observation requires controlling the experiment better than over-the-air. Measurements of PER vs SINR would be more compelling without random fading, unintended interference from other Wi-Fi devices, non-Wi-Fi interference, etc.

Another factor to consider would be differences in the waveforms transmitted by the interfering WARP and Wi-Fi devices. In a previous post you described the Wi-Fi devices as using "802.11a on 5GHz channel 3, bandwidth 40MHz with raw bitrates of up to 130Mbit/s". This must be an 11n waveform (11a doesn't support 40MHz or 130Mbps) right? The HTMF waveform structure has a 'midamble' (mid-preamble) which includes an HTSTF symbol. Maybe this symbol is re-triggering packet detection in the brief window after your code resets the Rx PHY following the SIGNAL (L-SIG for 11n) field processing? Attaching an oscilloscope to the OFDM PKT DET debug signal would be one way to see this.

running on 40 MHz bandwidth with 135 MBit/s to prevent triggering of packet events on the Nodes themselves

Important point here- an HT40 waveform will still trigger packet detection in the 802.11 ref design Rx PHY if either half of the HT40 waveform is centered in the same channel as the WARP node. The 40MHz HTMF waveform uses the legacy STF/LTF/SIGNAL in both halves of the spectrum in order to trigger reception and deferral by legacy nodes.