, , , , Audrey Puls
Tightly packaged receive systems pose a challenge for noise measurements. The only available receiver outputs are often user data: link benchmark or diagnostic information that can be scavenged with software. These could include data rate benchmark, self-estimated signal-to- noise ratio, or detection probability. Our "model" for their noise response may be a guess at an dependence on carrier-to-noise ratio (CNR). A general-purpose system-level noise measurement must satisfy these vague constraints while blind to implementation details in the receiver. Here, we propose a model and experiments to implement this blind measurement. Automated bench- top experiments establish a live communication link, and probe the user data response to varied signal attenuation and excess noise. We discuss the specification and design of the precision programmable excess noise source, and a highly directive link attenuator for testing transceivers. A new non-parametric regression procedure estimates the system noise (or noise figure) from the input conditions and summary statistics of the user data. Last, we estimate measurement uncertainty by combining traditional methods with a Monte Carlo method for random errors in the regression. Case studies apply the measurement to consumer IEEE 802.11 and GPS L1 equipment. These include verification by repeatability testing and cross-comparison against Y-factor measurements.
IEEE Transactions on Microwave Theory and Techniques