Meeting Topic: Sampling Rate Conversion - A Saturday afternoon discussion
Speaker Name: James D. (JJ) Johnston - AES & IEEE Fellow, and Bob Smith, Principal Scientist - Acoustic Systems, Physio-Control
Other business or activities at the meeting: Reminder of AES desk at the NAMM show.
Meeting Location: Redmond, WA USA
The PNW Section kicked off 2016 with a Saturday afternoon session about sampling rate conversion (SRC). AES/IEEE Fellow James D. (JJ) Johnston gave a tutorial on sampling, how sampling creates and controls the whole process, what a SRC does, how it works, and what it should and should not do. Bob Smith, of the AES PNW committee and an acoustic scientist at Physio-Control, then gave a presentation about SRC testing, with results showing aliasing and step response of different filters. Snack breaks were held about every hour. The meeting was held at a new, as-yet-to-be-named audio research company in Redmond, WA. 41 people attended, about 18 being AES members. Full A/V documentation will be available as well as the Powerpoint decks, demo files and scripts at the PNW Section meeting archive site:
JJ began with the Shannon Sampling theorem (i.e. Nyquist criterion) which dictates all with PCM digital audio. An SRC must never allow aliasing, where unfiltered spectrum images overlap the desired baseband audio signal, otherwise "stuff" happens. JJ played an example that shall never be reproduced again, but that did prove the point of "don't do that".
So, now you need to interpolate data? Yes, and to satisfy what the ear is sensitive to, including the frequency domain. He continued with problems with filter design and how that can interact with the original signal, which might have used half-band filters to bad effect. JJ has stats on thousands of CDs showing much questionable antialiasing (also much poor quantization, much clipping, and a whole host of other problems in various "don't do that" categories).
Continuing with filter design, JJ covered FIR, IIR, an attempt at apodizing, and half band filters, as well as filter calculation costs, efficiency, and multi-step filtering. Rate conversion at fractional rates, i.e. 8/3 fractional rates were introduced. Some mention of how to do rate reduction in steps was included, perhaps including a bit of confusion on how to do that (n.b. JJ added that comment). Finally, how to do conversion ratios with terrible least-common multiples was briefly discussed. FIR filter design was demonstrated using Matlab (expensive) or Octave (freeware!) showing ripple and pre-echo problems.
Throughout the discussion, the use of SRC for subsample delays was mentioned repeatedly, both for fixed and changing delays.
After a snack break, Bob Smith discussed SRC tests. He began with a photo analogy of incremental degradation - the photo gets blurrier with more aliasing.
Audio aliasing (spectra not filtered reflecting back into the passband) examples were played. JJ pointed out that this aliasing demo relates exactly to bad clipping scenarios during digital processing as well as for SRC.
Charts of step response (square waves) through different filter types (linear phase, minimum phase, hybrids) were shown, with different "ringing" at transitions. Such tests may help you determine what filtering is used in a DUT. Lastly, SRC example test result charts were shown, using the Octave script for testing.
Next, several door prizes were awarded - booth swag obtained from the 2015 AES 139th Convention in NYC, courtesy of Gary Louie:
Finally, JJ had a coda on SRC testing:
The message was that you should test your SRC yourself with your setup - Octave script provided. He showed example charts of a good and bad SRC, upsampling and downsampling.
Written By: Gary Louie