receiver sensitivity measurement

Hi all,


I am measuring the sensitivity of some radio cards, but I am unable to reach the expected sensitivity level. The sensitivity criteria is based on packet-error-rate, and I am using an RF shield box with the last 30-dB of the path loss inside the box in order to shield the card from interference. The sensitivity I'm getting is about 3 to 4 dB worse than I expect, and is consistent between samples.

I have tried adding an LNA with its own isolated power supply just before the radio card, but there is no improvement to the sensitivity. This suggests to me that the demodulator simply requires a higher SNR to demodulate than the datasheet suggests. Either that, or this is a multiplicative noise source. The radio card is hooked up to a PC via a PCIE interface, so perhaps somehow the PC's noise could be affecting the communications.

Any suggestions for how to debug this are welcome.


thanks,
Aaron

FYI: I spent a lot of time calibrating the path loss, so I'm reasonably confident that its not the issue...

- - - Updated - - -

thinking about it a bit more, a multiplicative noise source wouldn't explain what I'm seeing, because the radio is able to demodulate properly with adequate signal level. Also, if the noise was coming in directly to the radio card (from the PC for example), I would expect an LNA to improve the sensitivity. To me, it seems that either the demodulator is bad, or the noise is entering the channel (coaxial attenuators and cables) before any external LNA.

jeez where do I start?

Did you calibrate the signal power level accurately? I mean it woul be very easy to be off a few db with a stack of attenuators, and not precisely knowing the modulated transmit power.

It does not seem to be ingress, since you would leak in a modulated signal at a higher level, thru a cover ***** or something, than your attenuated modulated signal could be lower (as read on the attenuator) than the actual receiver input power is.

At very low signal levels, the reason a receiver fails to demodulate can be a number of things: thermal noise, a single low level spur, phase noise, etc....where as at a higher input signal level it might only be linearity and phase noise that matters. So I would go looking for these ghosts, like add a big value ceramic capacitor on the power supply leads to suppress a possible low level oscillation doe to lead lengths, and other things like that. Also, try to get an oscilloscope on the demodulated output....often the first bit or two of the header is corrupted due to timing/ringing/synthesizer settling time, and it is less of an issue at high SNRs.

I calibrated the signal power level. I will double check how accurate I was.

I intend to check out the power supply today. Actually the results are good compared to those reported online for the same application. The main thing is I cannot reach the datasheet level.

Bare in mind that this is a final product. I didn't design the radio chip, and I only have access to what I see on the PC. To some degree I can mess with the power supply a bit, but even that is difficult, as the radio is connected to the PC via an PCIE slot.


regards,
Aaron