Are there any methods of detecting, measuring, and/or estimating I/Q imbalance (from, say, an upstream quadrature mixer and/or ADC), without using or having a controlled test signal (other than whatever the RF antenna is currently picking up).

  • $\begingroup$ Receive or transmit path? $\endgroup$ – Marcus Müller May 12 '18 at 0:39
  • $\begingroup$ Receive. Transmit path is possibly a good topic for someone to ask a separate targeted question. $\endgroup$ – hotpaw2 May 12 '18 at 5:19
  • $\begingroup$ ah, by the way, the video I refer students to on the effects of IQ imbalance is this GRCon16 presentation by Matt Ettus; it's often a bit "symptomatic", but it does kind of allow one to follow around on how to link strange things happening to one's constellation diagram and the receiver imperfections (IQ imbalance somewhen after 20 minutes into the talk). $\endgroup$ – Marcus Müller May 12 '18 at 8:38


What is done in practice is simply disconnecting (or even better, terminating properly) the RF input of the receiver.

Then, you observe empiric noise variance separately on I and Q, and normalize so that they are equal.

This assumes that NF variations are benign, but it does work. If you can, you'd of course try to increase noise power.

USRP transceiver daughterboards can be RX IQ imbalance calibrated by letting TX transmit pseudowhite noise at a slight frequency offset (to average out potential TX imbalance) and observing the intra-board crosstalk. Same would work for any sufficiently random external signal. If you have a dominant external signal, intentionally tune a bit off, and average enough.

  • $\begingroup$ The stochastic approach is used to do two things: (1) match variance of I to Q; (2) eliminate cross-correlation between I and Q. These address gain and phase mismatch, respectively. This presumes that I and Q should have the same power and that they should be uncorrelated as is typically the case. It also typically presumes that the IQ imbalance is memoryless; however, if mismatch is caused by a difference in a baseband filter between I and Q for example, it will have frequency-dependent components. The described approach can still address this but needs a lot more data. $\endgroup$ – user35336 May 12 '18 at 16:22

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.