Junction Noise

[Grant Richter]

I was thinking in particular about the effect of the equal valued and binary
weighted summing resistors in the Buchla 266 "Source of Uncertainty".
The schematics are on my page at http://www.musicsynthesizer.com
(Grant Richter)

I am currently working on a new 1200 series module, based on the quantized
voltage source in the 266. I was trying to figure a way to extend into an
audio processor. One obvious improvement is to force all quantized steps to
83 mv (semitone) values so the module outputs calibrated 1v/oct scales.

My idea was to switch off the XOR feedback and use the shift register as a
transversal filter. Once again, probably easier to just try it than figure
it out mathematically.

BTW I put up a preliminary manual for the Joy Rider filter at
http://www.wiard.com

I still have seven filters left.

Best,

Grant

----------
>From: Martin Czech <czech at Micronas.Com>
>To: grichter at execpc.com
>Cc: synth-diy at node12b53.a2000.nl
>Subject: Re: Junction Noise
>Date: Thu, Sep 7, 2000, 3:58 AM
>

>
> :::Could someone give some general rules for "transversal" filters?
> :::If I am using the term correctly?
> :::
> :::This occurs in tapped binary shift registers where multiple taps are summed
> :::to get a multi-level output voltage.
> :::
> :::This also has frequency selective effects related to the number of taps,
> :::the tap weights and clocking frequency?
>
> Yes. Especially the later is interesting, it means Fc control via
> clock frequency.
>
> :::So it is a FIR filter? Short of going and buying some books on DSP, are
> :::there any useful rules of thumb, or handy references?
>
> This is a FIR filter, but a very special one, with fixed levels "1" and
> "0". But we could assume that we have a linear tapped delay line
> and a strange input signal of "1" and "0" instead. So normal
> FIR theory applies.
>
>
> Book: Rabiner & Gold. the holy bible of DSP.
>
> In "Art of Electronics" a nice 32 tap design is shown.
> The basic idea is that the tap weights are really the impulse response
> of the filter, and this is related to the frequency response via
> Monsieur Fourier's theory.
>
> The problem with FIR filters is that we only obtain zeros.
> That means: we need quite a lot of zeros to keep the response down
> in the stop band(s), i.e. filter length is usually > 64 or 128
> or so. OTOH FIR filters have wonderfull properties, eg. phase
> linearity. All partials traveling through such a filter will arrive at
> the same time.
> And: there are optimal filters, ie. a theory that prooves that no other
> FIR filter with given length will fit better to a given spec
> in some sense.
>
> I've toyed arround with such filter optimisation software
> (Remez exchange algorithm).
> If you want, I could easily compute some impulse responses/
> frequency responses.
> If think low pass and band pass are especially usefull.
>
> Even if we have no shift register feedback at all, sampling
> analog noise and filtering it could still be usefull.
> This time alias and qunatisation noise is very much wellcome!
>
> m.c.
>
>