I am currently trying to measure the latency through my alsa application. I will briefly explain my application and set-up. The application has a test harness which sends audio data to my application. To measure the latency the test harness reads in a .wav file(one which I have generated. This contains a short pulse of a sine wave. The rest being zero data.) and also drives the serial port at the point were it detects the sine pulse. As previously explained the audio data is passed through my application and the alsa driver. The difference then between the start of the sine pulse and the output from the sound card is the latency through the system. I measure this with an oscilloscope. Also just to note the test harness is driven from the same sound card as the application, so there should be no difference clocks.

Now that I have explained the test scenario, I will explain the behaviour that I am experiencing. To start off, starting the test harness at different instances revealed that the latency measurement is different by maybe 20ms or so, although the alsa buffer levels indicate all is correct. Once started the behaviour is such that the difference drifts ever so slightly. After around 10 minutes the output flips to the beginning and the process starts again. This happens continuously.

Without establishing the cause of this problem will ultimately mean the measurement is useless. Could this issue be related to the serial port, alsa buffering, kernel scheduling etc. Can somebody with the relevant experience help. Please.