WARP Project Forums - Wireless Open-Access Research Platform

For any given receive power at the antenna connector, there exists a (Rx RF gain, Rx BB gain) tuple that minimizes receive EVM (i.e. maximizes SNR). AGC attempts to do this automatically based on the EVM curves in the MAX2829 datasheet. So yes, leaving AGC enabled is probably a good first step. Note that it is possible for the AGC to incorrectly guess appropriate gains; this might skew your BER/PER results. Another approach is to manually tweak Rx RF gain and Rx BB gains for each SNR point along your x-axis of your experiment, but this could be pretty tedious at best and impossible at worst (e.g., a fading channel with a fluctuating instantaneous SNR must be tracked with AGC).

As far as transmit gain is concerned, you could use Tx BB gain and Tx RF gain to effectively adjust the average SNR at the receiver. Alternatively, you could leave the Tx gains alone and adjust SNR by using a variable attenuator. That's how we performed this throughput characterization of the 802.11 Reference Design.

PatC422 wrote:

If I kept the Tx BB at 3 and gradually stepped down the Tx RF while not using any AGC and keeping the Rx RF and Rx BB constant, could you think of any issues with such an approach?

The actual values of SNR are not critical as long as they are seen to decrease more or less incrementally because I can calculate them based on the fact that I intermittently transit zeros to calculate the noise variance at the Rx. As a result, I can then place calculated SNRs, (var{Rx_signal}/var{Rx_signal_zeros}) - 1, into simulation for comparison. I also intend to use a simple LOS channel with not much fading as I am more interested in characterising the behavior of the signal processing rather than the channel.

I realise that I will probably have to 'sound the hardware' too by using a known test signal on a transmission line and 20 dB attenuator in order to quantify the effects of power imbalances, etc, in the RF chains.

Any further comments & discussion would be most appreciated.

You'd see degrading performance as a function of decreasing Tx gain, but much of that performance degradation would come from quantization at the Rx ADCs due to improper Rx gain choices rather than any "fundamental" receiver-noise-based SNR degradation. That's not typically the kind of signal degradation people talk about when they say "low SNR."

Also, you might want to take a look at this paper with regards to calculating SNR. You can directly compute SNR by looking at the EVM of your received constellation and don't need to rely on any intermittent zeros in your receive vector. This measurement approach would also be the same regardless of what Rx gains are used, so AGC can be a pretty natural inclusion.

Hi again,

If I may be permitted to ask another question. I was wondering what are the shortcomings of WARPv2 over WARPv3? I am using WARP v2 and I would just like to be aware of any limitations.

Cheers,
Pat

There are a number of hardware differences between WARP v3 and WARP v2 as Patrick mentioned.

The major difference is that WARPv2 uses a Virtex 4 FPGA while WARPv3 uses a Virtex 6 FPGA.  This drives different FPGA architectures due to the support for different peripheral sets by each FPGA (look at WARPLab 7.5.1 for a comparison of the two FPGA architectures for the same WARPLab release).  This architecture difference is why there are differences in supported sample buffer sizes.  There is also a minor difference in timing for Ethernet triggers, since we are not able to "snoop" the Ethernet stream in WARPv2 due to a different Ethernet controller (see an explanation of Ethernet trigger timing here).

I am honestly not well-versed in this work, so I'm not comfortable attempting to provide insights on specifics. But I can link you to a couple of other papers that might help. I think this paper also makes the assertion that SNR = 1 / AEVMS (which is consistent with your comment). It cites this paper for that assertion.

It's hard to use RSSI to get an estimate of noise power for the denominator of your SNR term. Check out the RX RSSI OUTPUT vs. INPUT POWER graph on page 19 of the MAX2829 transceiver datasheet. The RSSI output voltage levels off and flattens when the input power is low (which is the scenario you are describing if you aren't transmitting anything). I don't think its safe to conclude anything about very low powers using RSSI. I don't have any personal experience, but I think you are going to find it much easier going with the AEVMS approach.

Yeah, I think that's what that phrase is referring to. I've colloquially used "pseudo-EVM" before to describe the same thing -- the Euclidian distance between the received symbol and the symbol that we estimate that it actually was (recognizing that decoding attempt may be in error). Usually this subtlety doesn't matter in experiments with tools like WARPLab since you have the luxury of knowing exactly what was sent.