Conference napalm::commusic_v1

I find that I do not miss analog functions, except long-term memory.
I use a multi-track tape to record tracks, and later combine them onto a 
cassette.  Doing that digitally would be very expensive.

I have a high-performance digital signal processor, though it is much less 
flexible than your description of analog equipment.  My high-performance
digital signal processor is a Yamaha DX7.
    John Sauter

	John,
	   I'm confused about your reference to your DX7 as
	a digital signal processor. I had thought that it
	was FM synthesis only. I would define DSP as stated
	in the base note as digital manipulation of analog
	sourced signals, like...

	source -> A/D -> digital processing -> D/A -> output


	TOm,

	   A real high performance digital signal processor
	wouldn't need to be dedicated to one function. It 
	would be able to perform as many calculations as 
	needed (efficient code permitting, and FAST processor).
	   After all digital processing is using the same
	mathematical manipulations that your analogue modules
	are doing.
	   For instance balance modulation is a four quadrant
	multiplication in the amplitude domain, easy for a processor.
	   Filtering is done with multiplication in the time
	domain upon time slices of the signal.
	   The digital processing is limited only by the 
	multiplication time for your CPU (+system speed as a whole
	I/O, memory fetch etc...) and the intensity of calculation.

re: .2--By your standards the DX7 is not a digital signal processor, since
it does not have any A/D converters.  However, in my opinion, if a "signal"
can be produced by purely digital means, no A/D converter is needed in
order to process it.  FM Synthesis is a processing technique, as is
Additive Synthesis.  The DX7 can do these things, and manipulate the result
based on its envelope generator and LFO signal.  To me, that is enough to
make it a digital signal processor, even though it has no A/D converter,
and lacks a lot of the flexibility of other digital signal processors. 
    John Sauter

re 2
Yes they would too have to be dedicated.  I have done all the calculations
and preliminary system level design.  Even TTL processors running top speed
could not do two functions on separate signals.  It's just too much to ask.
Tom

(OK, so this note is wandering a little from the original question...)

As to the current question of whether or not a digital signal processor would
have to be dedicated to a given function, I think not.  We already have the
DMX-1100 (I think that's the right product number) which is exactly what is
being discussed, a general-purpose programmable digital signal processor.  If
that's not enough, Cromemco is offering a product for the S-100 bus called the
'Maximizer coprocessor subsystem'.  In a nutshell, this is a FAST bit-sliced ECL
processor capable of about 12 MIPS!  It has a 48 bit microcode word length with
16K bytes of loadable dual-port memory.  Put a A/D-D/A board on the bus and,
voila, a DSP.  Granted, a pretty expensive one (the Maximizer costs $3495), but
then the DMX cost about $10,000 last time I looked (years ago). 

A related question might be "Yes it's possible, but are DSPs cost-effective?"
I think the answer is still no, unless you have such varied or esoteric signal
processing requirements that a programmable DSP is the only thing that is as
flexible as you need.

Another way to attck the problem, would be to use a hybrid approach, analog
gear under CPU control.  Sometimes this turns out to be pretty easy, sometimes
not.  The Emu analog chips are supposed to be nice, but hard to CPU control,
since they require analog control voltages.  (By the way, do those chips still
exist?)

It looks like it might be pretty easy to build computer-controlled filters
since National Semiconductor has two good filter chips, the MF10 dual two-pole
switched capacitor filter and the MF4 4th order siwtched LPF.  I have some MF10s
at home (Radio Shack was selling them a while ago) and have ordered some MF4s.
First I need to put LPFs on the outputs of my Casheab synthesizers (see my note
early in this file if you don't know what these are, I have 4).  Then I'm
probably going to build a speech board and I need another LPF for that.  Then
I'll look at building some general-purpose filters under CPU control.  This
shouldn't be impossible since both the MF10's and the MF4's cutoff frequencies
are controlled by an external oscillator.  So to computer-control them, you
need to build a digitally-controlled oscillator.  There are abviously LOTS of
ways to do that.

Well, to answer TOm's *original* question "do any of us miss analog signal
processing gear?", my answer is "Youbetcha!".  I only have two pieces of
outboard signal processing gear currently (leaving the DX7 out of it), a digital
delay line and a Fostek (spring/analog) stereo reverb.  I would like some other
analog gear, a partial list being:

    1) some outboard filters (LPF, BPF, HPF)
    2) at least one ring modulator
    3) at least one good filter bank (parametric)
    4) a fuzz box (when you want *that* sound, nothing else will do)
    5) a vocoder
    6) some envelope followers and an A/D board to input them to my CPU
    7) a 16/4 analog mixer under CPU control

I think my need for some more outboard analog gear stems from the fact that I
started out in the late 60s/early 70s in a pretty traditional (?) electronic
music studio with some old signal processing gear (from a physics lab) and a
big Moog. There were sounds in the Moog that I know I can't get with my DX7;
I'd need some of the analog gear above and some MIDI code running in my CPU
before I could reproduce them. 

While on the subject of analog gear, I noticed that PAIA has come out with a
vocoder kit for $100.  If anyone has anymore information on it, please start
another note and tell us about it.

-jim

I bought the vocoder.  When I build it I'll let you know.
Maybe hear it over the phone.
TOM

	   Could you describe it a bit for those of us who
	don't get their catalogues anymore???
	  I would like to build one eventually. I've thought
	about converting a graphic equalizer I've got into one.
	Or using the new MF chips or the new 7? band graphic
	equalizer on a chip. But I havn't done anything yet.
	
	   How many band does it have. Does it use the 3900
	quad op-amps?? or better ones??

	   Q: Could anyone critique the Roland vocoder??
	      Like is it worth the money and how much is it$$.


	John

This might not be the PAIA vocoder, but is an example for
those who are unsure what we're talking about.

TALKING SIGNAL-----------I
                         I
    -------------------------------------------------------
    I          I          I          I          I          I      
********** ********** ********** ********** ********** ********** 
*        * *        * *        * *        * *        * *        * 
* filter * * filter * * filter * * filter * * filter * * filter * 
*  c-eb1 * *   e-g1 * *  g#-b1 * *  c-eb2 * *   e-g2 * *  g#-b2 * 
********** ********** ********** ********** ********** ********** 
    I          I          I          I          I          I      
********** ********** ********** ********** ********** ********** 
*envelope* *envelope* *envelope* *envelope* *envelope* *envelope* 
*follower* *follower* *follower* *follower* *follower* *follower* 
*        * *        * *        * *        * *        * *        * 
********** ********** ********** ********** ********** ********** 
    I          I          I          I          I          I      
    V          V          V          V          V          V      
control inputs  
********** ********** ********** ********** ********** ********** 
*        * *        * *        * *        * *        * *        * 
*  VCA   * *  VCA   * *  VCA   * *  VCA   * *  VCA   * *  VCA   * >summed out
*        * *        * *        * *        * *        * *        *  finished
********** ********** ********** ********** ********** ********** 
    ^          ^          ^          ^          ^          ^      
    I  signal  I  inputs  I          I          I          I      
********** ********** ********** ********** ********** ********** 
*        * *        * *        * *        * *        * *        * 
* filter * * filter * * filter * * filter * * filter * * filter * 
*  c-eb1 * *   e-g1 * *  g#-b1 * *  c-eb2 * *   e-g2 * *  g#-b2 * 
********** ********** ********** ********** ********** ********** 
    I          I          I          I          I          I      
    -------------------------------------------------------
                            I
Chords played on synclavier-I

TOm

OK let's see...

	use the digital recording standard ca. 44kHz.
	one sample must be processed in no more than 22 us.
	@ 12MIPS it can do ca. 250 instructions per sample.
	a typical simple filter has about 50 instructions.
	more complex functions are worse.
	Assume any ADC arrangement necessary (including muxing)
	to get necessary throughput.
	Maybe you could filter five separate signals.
	More complex functions would monopolize the processor and memory.
	
	Sure, a signal processor could do lots of different things, but
	probably best only one signal at a time, maybe more than one 
	function on that one signal.

TOm

And generating signals is far easier than processing incoming.
TOm

	   There are chips that can do 16x16 multiplies in
	50ns plus 35ns static RAMS for scratch pads.
	Sooooo 256 multiplications @50ns is....
	256 samples in 12.8 micro-secs.

	Coupled with Bit-slice processors and dedicated
	microprograming intended to realize the above,
	sounds a little bit faster....

	So your simple filter goes down to 2.5ms per function.
	TTL is NOT the way to go if you want to get anything
	done, it may be fine for some things but not for this.
	
	John

I was using RAVAN's numbers on the Cromenco.
Anyone have performance figures on LucasFilm's ECL sound processing
system? set up by none other than what's-his-name himself.

Radar imaging, anyone?
TOm

BTW, I am the standard ECL test group.
TOm

	O.k. I give up......but I
	am going to build it anyway.

	John

Define BTW = "By the Way"

Let's talk GaAs.  Now on sale at Radio Shack, Picologic from
Gigabit in Thousand Oaks California; typical gate delay bonding pad to bonding 
pad 75picoseconds (or 75000 fempto-seconds).  Build hybrid circuits to avoid 
PC board runs.  Wild guess a multiplier is 35 gate delays, you get
2.6 ns multiply.  With flip flops (now on sale ) with a cycle time of
333ps, and 10 gate delays decode, that's 750+333 is about 1 ns access.
so you get a 2.6 ns instruction.  make it 2.5, you get line delays of about
2 nanoseconds maybe closer to 4, it's 7.5ns cycle.  That's 133MHz 
instruction speed.  50 instructions gives us 2.6 MHz Band Width.
Audio is 40 kHz BW, so that's 65 separate filters.

Japanese Josephson-Junction cryo-logic gate arrays might triple or
quadruple that.  If you have a warm coat.

Let's talk optical logic and Fabry-Perot interferometers.  1 pico-second 
(1000 femto-seconds, or 1000000 atto-seconds) gate.  With similarly fast
memory, you get 200MHz BW, or 4875 separate filters.  That memory is
not yet on the distributor's shelf.

Now what if we use...

whew! I'm worn out.

TOm

	The following is in a pseudo-japanese dialect...

	"You amelican lunning-dog impeeliel-rar-monging pigs!
	 Take dat yankee-doorle-dandy pig-dogs! I no die! I no
	 die Joe! You die, I teach you die, Joe!"

	Signe Hiye!
	Joe
	Alias China-joe
	Chop-Chop!

Solid State Circuits 12-84 p 1038 correspondance reports a 267ps gate array
with 5000 gates.

Last night UPS gave me my PAiA vocoder kit.  It came in a funny equilateral
prism cardboard box.  I have built it half way.

As you can see in the catalogue, it is for their larger cabs, about
17 3/4" wide, 1 3/4" high, a few inches deep.  The single-sided PCB
is to be mounted to 3/16" aluminum front panel, like other PAiA modules.
Thus, it has no box, no power supply.  The movers forgot to lose my
+-15 V power supply, and I will use this for check out and initial use.
There are 9 I.C.'s, about 80 caps, about 60 resistors.  It is not a hard
kit.  I am missing 4 1nF caps, and will look for them at a store at lunch.
Maybe working tonight.  I still have to install 12 caps as well as the jumpers,
and build the front panel.

There was an etch missing, and an extra sheet with the kit instructs me how to 
fix it.

It has two banks of eight filters, but I have not seen what their range is.
It takes a mic in.
If I build it right and it works, perhaps I will accept tapes of your own 
sounds with self-addressed stamped envelopes to show you how it sounds.

This is a device that impresses the filtering characteristics of the vocal
tract of the person talking into it onto any other audio input.  People use
it to make a synth into a cooperative back-up vocal group.  It is also
possible to send an electronic drum into the vocal input and impress it's
power vs. frequency characteristics on the synth or other sound.

Tom Janzen DEC 150 Locke, Marlboro MA

Wed 13-Mar-1985 08:57 EST

You should note that the PAiA 6710 vocoder kit is not step by step for a 
beginner.  For example, the instruction says,"install all the sockets."
then "install all the resistors"., so then you look at the parts list
for the resistors and put them in the indicated places.  If you have built
a couple other kits from Heath or PAiA you can probably handle it fine.
Tom

It works.  I do not have any experience with vocoders, but have heard them
on radio science-fiction shows ("Ruby") and used by Laurie Anderson.
It seems to be about the same quality of sound.

problems:

the four 1 nano-farad polystyrene caps were missing.

Screws were too short I felt for the switches that didn't have threaded holes.

I had to cut off the nibs off the potentiometers to fit them to the panel.

There no mounting holes in the panel (although mounting to the PCB is shown
	in an illustration)

Many of the jumpers' lengths, given in the text, seemed to be in centimeters
	rather than inches; they are very wrong.

The etch was a little loose and easy to detach when it's hot.

The 5" inch bare wire on page 8 is really supposed to be 3.5 inches.
The wire on AU is a little short.

Tom

maybe it's just me, but I couldn't hear the vocoder effect on that last
posting :-)

	Eirikur