# AutoCorrelation of 2 waveforms to identify noise patterns

I have 2 audio recordings of the same sound, one with noise and the other without:

(This is the Opening Ceremony from Epyx Winter Games on a vintage Macintosh SE, for those who are curious. It was recorded from the headphone jack using a Sony D100 recorder at 48kHz 16-bit. The Mac generates sound at 22kHz 8-bit.

The audio with noise is caused by a 16MHz 68000 based accelerator card. The audio without noise is with the accelerator disabled, running off the stock 8MHz 68000 CPU. I am trying to use Audacity to determine if the noise is recurring at fixed intervals like every 15ms or so. The thinking is that if the noise is on a fix recurring pattern, the root cause of the noise might be better understood.

I am aware that Audacity provides a Correlation feature, but I have no experience with it, nor do I fully understand it, hence my post here. Specifically…

Analyze > Plot Spectrum… > Algorithm > Standard Autocorrelation

If I put both of the aforementioned tracks into the same Audacity window and Select All and then choose Plot Spectrum and then Standard Autocorrelation, I see a waveform. But I am not understanding the Autocorrelation feature well enough to understand how to determine if the audible noise is recurring every fixed number of milliseconds.

Any advice you can share on how to achieve my aim would be greatly appreciated.

Thank you.

The track spectrogram view reveals quite a lot about the noise (https://manual.audacityteam.org/man/spectrogram_view.html)

Thank you for your kind suggestion. Changing the Spectrogram settings to grayscale makes the vertical bands even easier to see. Doing a hand drag measurement between two adjacent bands indicates they are 16ms apart. That is useful information.

Interesting that it’s 16mS apart.

You said the clock was 16 MHz which has a cycle time of 62.5nS and this multiplied by 256000 = 16mS.
Can’t be a coincidence.

except that the pulse frequency looks to be around 16 1/2 ms to me (taking the average of 10 pulses).

Could very well be that the last clock pulse (after a certain division ratio) triggers a reset for a counter and that in turn has a certain delay time.
My guess of /256000 was only based on the figures quoted.
All sound cards take the master clock and divide by a certain ratio and use that.
Then after a certain amount of pulses, they present the ADC or DAC data to some bus which also takes a finite amount of time.

Perhaps a simpler explanation (conjecture): 16 1/2 ms is very close to 60 Hz, which is the mains AC frequency in many countries.

Yep, another very big possibility, well spotted, I didn’t even think of that.

In regards to the timing now being discussed (16.5ms or 60Hz), I have been told there is a “vblank” interrupt on the Macintosh SE which occurs every frame. The Mac redraws the mouse arrow pointer and updates the sound buffer. The sound buffer is stored in a fixed location in RAM and contains 370 8-bit samples. The sound system is tied to the video system, and outputs one new sample for every vertical line scanned to CRT. There are 370 samples in the buffer because there are 370 vertical lines on the screen (342 visible and 28 for retrace). Audio plays fine when the Mac runs off its stock 8MHz 68000 CPU, but it hasn’t yet been determined why there are audio glitches when 16MHz 68000 accelerator is enabled.

Thank you again for kindly offering your thoughts and suggestions on how to use Audacity to track recurring noise patterns.