What resampling is being used here?

[Kraku]

I recorded a 44.1 kHz signal from computer into my E-mu e6400 Ultra. I had set E-mu to sample the signal with the same 44.1 kHz samplerate.

When I recorded E-mu playing that signal 2 octaves down, I got the result seen in the attached image. There should be nothing above 5 kHz in that picture, yet there's clearly some extra stuff there.

Can anyone recognize the used resampling method from these clues?

[mystran]

You can try to estimate the sidelobes and then based an educated guess based on that, but this spectrum is quite noisy (a test signal would be better) and logarithmic frequency scale makes it very hard to see what's going on at the high frequencies.

That said, linear interpolation gives a little over 24dB attenuation at the peak of first side-lobe and if you eyeball about 24dB up from the top of the 10kHz bump, we're pretty close to the spectrum at ~3kHz so based on this plot alone I'd guess linear. Catmull-Rom should perform a little bit better than this, but I guess some other cubic is theoretically possible too... but like probably it's just linear.

[juha_p]

By E-mu e6400 technical documentation:
...
G-chip Sound Engine (2.0)
EMU8000 Chip
H-chip Digital Filter (1.5)
...

If you google search with above terms you may find the info you're after... .

[Kraku]

@juha_p: the documentation I was able to find didn't reveal the information.

@mystran: I probably need to make a test with a sine wave and see what comes out of it.

[hugoderwolf]

You probably won't see much with a sine either. Maybe try a square wave with integer sample period?

[mystran]

@juha_p: the documentation I was able to find didn't reveal the information.

@mystran: I probably need to make a test with a sine wave and see what comes out of it.

Sinewaves are not generally very helpful test signals for anything other than estimation of harmonic distortion and even for those you really need a chirp and then you go through additional hoops to actually interpret the results.

My two cents is that your number one test signal for everything should be a good quality bandlimited saw-wave and your number two test signal should be a good quality bandlimited pulse wave where you can adjust the pulse-width. You want a predictable spectrum you can compare with reference, not a single isolated spike.

[juha_p]

@juha_p: the documentation I was able to find didn't reveal the information.

...

D. Rossum (was one of four central peoples of E-mu Systems back then) has lots of patents for Creative/EMU and in a 4 part interview he mentions: "Internally, the EMU8000 was called the “G0.5” because it used a cost reduced G-chip algorithm for pitch shifting." and in part three: "The G in G-chip was taken from one of the authors of the technical paper that introduced us to high order interpolation – Phil Gossett and Julius Smith – and we called the algorithm “Gossett-Smith.”".
So, maybe "order seven" G-chip interpolator; US patent no. 5,111,727, David Rossum in background.

This (PDF) paper is referenced in many discussions: and there's a note 71 (p. 769) mentioned:

The E-mu Proteus sampling music synthesizer (1989) and
its relatives all employ seventh-order interpolation polynomials.
They are not exactly "Lagrange" though. They use a technique
in which a Remez exchange is applied to an "ideal" filter response
similar to that of Lagrange, but having lower maxima in the
stopband. This gives the deep notches advantageous in the La-
grange approach, but also the superior stopband rejection of a
sinc-based design. For more information, see the U.S. patent
on the fundamental E-mu G-chip interpolator; no. 5,111,727.
David Rossum.

Hmm... maybe something like this implementation.

[Kraku]

@juha_p:
The patent seems to describe the resampling algorithm in detail. Thanks!