Maker Pro
Maker Pro

Crosspoint Analog Switches

C

Charmed Snark

Jan 1, 1970
0
I'm trying to come up with a microprocessor solution to
replace the traditional analog synth patch cords, using an
analog crosspoint switch.

The idea is to replicate the mechanical matrix used in this
synth, but under microprocessor control:

http://www.vintagesynth.com/misc/vcs3.php

Google yielded some crosspoint video (analog) switch chips
with memory (latching). I came across some chips like the old
74HC22106 and the newer ADG2188 etc.

These almost solve the issue except that there is no 10k
isolation resistor (say) at the crosspoints. You could add 10k
resistors to the inputs (say columns) but if one input was
patched to more than one output line, then those two outputs
would be only isolated by about 50 ohms. So cross talk would
be a problem.

So my question is this- are there any crosspoint switch
solutions that include an isolation resistor at each
crosspoint?

The isolation resistor at each crosspoint allows each patch to
automatically be summed (mixed) like the typical summing opamp
circuit. It will also provide some level of isolation between
each line.

Warren
 
J

Joerg

Jan 1, 1970
0
Charmed said:
I'm trying to come up with a microprocessor solution to
replace the traditional analog synth patch cords, using an
analog crosspoint switch.

The idea is to replicate the mechanical matrix used in this
synth, but under microprocessor control:

http://www.vintagesynth.com/misc/vcs3.php

Google yielded some crosspoint video (analog) switch chips
with memory (latching). I came across some chips like the old
74HC22106 and the newer ADG2188 etc.

These almost solve the issue except that there is no 10k
isolation resistor (say) at the crosspoints. You could add 10k
resistors to the inputs (say columns) but if one input was
patched to more than one output line, then those two outputs
would be only isolated by about 50 ohms. So cross talk would
be a problem.

Can you sketch up a schematic? If the input goes to two outputs, why
would they need to be isolated if they now carry the same signal?

So my question is this- are there any crosspoint switch
solutions that include an isolation resistor at each
crosspoint?

The isolation resistor at each crosspoint allows each patch to
automatically be summed (mixed) like the typical summing opamp
circuit. It will also provide some level of isolation between
each line.

I've done several such muxes, all in the frequency range up to 15MHz. Up
there isolation is much more difficult. The standard trick is this: Run
anything across the X-point chips strictly current mode. No voltage
summing, always currents.
 
W

Warren

Jan 1, 1970
0
Michael A. Terrell expounded in
Have you looked at the switches from Maxim IC? They
have a whole
series designed for studio routers.

I have looked at the MAX4456, but I suspect that is very old
now (hard to get at the moment). I'll go back and look at
their line up. I just googled for "crosspoint switch(es)" when
I came across MAX4456.

If they have "studio routers", that may be what I'm looking
for.

Warren
 
W

Warren

Jan 1, 1970
0
Joerg expounded in
Can you sketch up a schematic? If the input goes to two
outputs, why would they need to be isolated if they now
carry the same signal?

In analog synths, you want the signals to "mix" (not take
over). But if one signal is going to two destinations (or
more), you don't want the two different destinations coupling
tightly (the 2nd output may have other signals mixed in).

Imagine a row column matrix. In the mechanical plug version of
this, you plug in a plug that adds a 10k resistor between that
column Y5 and row X3. This replaces the need for a patch cord
between Y5 and X3.

If you then add another plug from col Y5 (input) to row X4
(output) there is now only a 20k coupling between X3 and X4,
while Y5 connects to both.

Further complicate this by adding a column Y6 input to be
mixed in with X4 (only). Y6 and Y5 are also now 20k isolated,
yet X4 can sum both inputs Y5 and Y6. (picture your typical
summer opamp circuit, except that the input resistors are
added at the matrix crosspoints)

In the ss chipped version, there is a connection made with a
very low (50 ohmish) resistance connection between row and
column (vs the 10k in the plug matrix version). This is ok,
when the column is going to _one_ destination (video), but not
when multiple destinations are involved.

The same scenario in the chip would result in only about 100
ohms of isolation between X3 and X4 (when Y5 was enabled to
both X3 and X4). This is almost equivalent to soldering Y5
directly with both outputs. Add another connection from Y6 to
X4 (only), and you might as well solder all connections Y4,
Y5, X3 and X4 together (which is highly suboptimal).

Yet in the mechanical plug version, X3 and X4 have some level
of isolation (20k) to allow for different output signals. Y5
and Y6 also keep at least 20k worth of isolation from each
other. This is an oversimplification with a complex network,
but you get the idea hopefully.

Just scribble out what I've described (just the matrix) and I
think you'll get what I'm after (when a crosspoint is enabled,
it is connected by a 10k resistor). The basic problem is that
the video switch is probably intended to not mix, but simply
to "route" from one input to one output.

But here, I'm looking for a way to route and mix, albeit with
limited isolation.

Warren.
 
W

Warren

Jan 1, 1970
0
Jon Elson expounded in
I've used the CD22M3494 with good results. It is really a
video switch, and I have no idea whether there might be
audible transients when the switch settings are changed,
but it worked well in the application we used it in.

It has the same limitation I mentioned earlier, but--

It does have the advantage of low price and _wider_ width
(16x8). I could adapt something like this to work with some
limitations.

Using only one chip, I could pair inputs with 10k resistors
going _in_ for each leg of the pair (sig + 10k into X0, and
same sig + 10k going into X1 for example). That way any pair
of the same signals would still wind up with 20k ohm isolation
between two selected outputs.

Given the low cost (and I like that they have DIPs), I could
add another, allowing 32x8 => one input could fan out through
4 different X0 to X3, each with their own isolating 10k
resistors. In this manner, I could mix up to 4 of the same
inputs to 4 different destinations, with 20k isolation.

Thanks for the chip reference. I'm going to look at this
possibility.

Warren.
 
J

Joerg

Jan 1, 1970
0
Warren said:
Joerg expounded in

In analog synths, you want the signals to "mix" (not take
over). But if one signal is going to two destinations (or
more), you don't want the two different destinations coupling
tightly (the 2nd output may have other signals mixed in).

Imagine a row column matrix. In the mechanical plug version of
this, you plug in a plug that adds a 10k resistor between that
column Y5 and row X3. This replaces the need for a patch cord
between Y5 and X3.

If you then add another plug from col Y5 (input) to row X4
(output) there is now only a 20k coupling between X3 and X4,
while Y5 connects to both.

Further complicate this by adding a column Y6 input to be
mixed in with X4 (only). Y6 and Y5 are also now 20k isolated,
yet X4 can sum both inputs Y5 and Y6. (picture your typical
summer opamp circuit, except that the input resistors are
added at the matrix crosspoints)

In the ss chipped version, there is a connection made with a
very low (50 ohmish) resistance connection between row and
column (vs the 10k in the plug matrix version). This is ok,
when the column is going to _one_ destination (video), but not
when multiple destinations are involved.

The same scenario in the chip would result in only about 100
ohms of isolation between X3 and X4 (when Y5 was enabled to
both X3 and X4). This is almost equivalent to soldering Y5
directly with both outputs. Add another connection from Y6 to
X4 (only), and you might as well solder all connections Y4,
Y5, X3 and X4 together (which is highly suboptimal).

Yet in the mechanical plug version, X3 and X4 have some level
of isolation (20k) to allow for different output signals. Y5
and Y6 also keep at least 20k worth of isolation from each
other. This is an oversimplification with a complex network,
but you get the idea hopefully.

Just scribble out what I've described (just the matrix) and I
think you'll get what I'm after (when a crosspoint is enabled,
it is connected by a 10k resistor). The basic problem is that
the video switch is probably intended to not mix, but simply
to "route" from one input to one output.

But here, I'm looking for a way to route and mix, albeit with
limited isolation.


Ok ... but ... unless I am having a brain freeze here:

Suppose you send something from Y5 and Y6 into X4. Y5 also goes to X3,
which happens to also get a feed from some other place, and so on. If
you feed the mux strictly current mode and make sure you have current
receivers on all X ports then the impedance at all Y ports is very high
(ideally infinity). Yet at X3, X4 and the others it is zero. Like a dead
short. So how could there be any continuing leakage for example from
whatever else feeds into X3 via Y5 and then into X4?

The zero ohms receiver is often also called a "transimpedance amplifier"
or TIA. The "economy class version" of that would be the emitter of a
common-base transistor stage. What you do have to adjust for can be done
on the drive side: If you feed two X-outputs from the same Y-input then
the current splits up 50:50. So does the amplitude. If you feed three
then 33%, and so on. So you'd have to up the drive signal accordingly.
 
S

Syd Rumpo

Jan 1, 1970
0
I'm trying to come up with a microprocessor solution to
replace the traditional analog synth patch cords, using an
analog crosspoint switch.

Don't know if it helps, but look at AD75019 - it's still available and
is a IIRC +/-12V 16x16 analog crosspoint. Annoying part is serial
addressing, but a slave PIC would do this easily.

I've used this in the past with great success.

Cheers
 
N

Nico Coesel

Jan 1, 1970
0
Warren said:
Jon Elson expounded in


It has the same limitation I mentioned earlier, but--

It does have the advantage of low price and _wider_ width
(16x8). I could adapt something like this to work with some
limitations.

The part number rings a bell. Just beware of the very odd numbering of
the rows and columns.
 
W

Warren

Jan 1, 1970
0
Joerg expounded in
Warren said:
Joerg expounded in news:[email protected]:
[Warren] wrote:
I'm trying to come up with a microprocessor solution to
replace the traditional analog synth patch cords, using an
analog crosspoint switch.

The idea is to replicate the mechanical matrix used in
this synth, but under microprocessor control:
...
Yet in the mechanical plug version, X3 and X4 have some level
of isolation (20k) to allow for different output signals. Y5
and Y6 also keep at least 20k worth of isolation from each
other. This is an oversimplification with a complex network,
but you get the idea hopefully. ...
But here, I'm looking for a way to route and mix, albeit with
limited isolation.

Ok ... but ... unless I am having a brain freeze here:

Suppose you send something from Y5 and Y6 into X4. Y5 also goes to
X3, which happens to also get a feed from some other place, and so
on. If you feed the mux strictly current mode and make sure you have
current receivers on all X ports then the impedance at all Y ports is
very high (ideally infinity). Yet at X3, X4 and the others it is
zero. Like a dead short. So how could there be any continuing leakage
for example from whatever else feeds into X3 via Y5 and then into X4?

Ok, using the Y5, Y6, X3 and X4 example, let's look at this from the
mechanical plug board pov first, that I am trying to replicate in
silicon. Remember each plug has a resistor in it for isolation.

Y5 is connected to X3 and X4
Y6 is connected to X4 only

The plugs are put in at the crosspoints, each of them containing a
resistor (I've assumed 10k ohms) :

(switch to fixed fonts here)

Y5 Y6
| |
| |
--|-----+--|-------- X3
+R=10k+ |
| |
| |
--|-----+--|-----+-- X4
+R=10k+ +R=10k+
| |
| |

This small matrix patch configuration is equivalent to:

Y5--+-R=10k--------X3
|
+-R=10k---+
|
Y6----R=10k---+----X4

From this, it is very easy to see that the isolation from X3 and X4 is
20k and also that Y5 and Y6 is also isolated by 20k. Y6 to X3 isolation
isn't great, but it is 30k ohms.

But when you use the silicon matrix solutions, the R is lowered to
roughly 50 ohms (or less). So then X3 and X4 is only isolated by 100 ohms
and likewise Y5 and Y6 is also only isolated by 100 ohms (like I said
earlier). This is completely unsuitable for voltage based signals (which
these are). Further Y6 is only 150 ohms away from X3.

What I think you're suggesting is to convert from a voltage based signal
and work with currents instead. The problem is that it must also accept
DC control signals (inputs are not always strictly ac signal). To apply
two conversions (voltage -> current and back) would introduce too much DC
error. Pitch control for example, is very fussy.

Warren
 
W

Warren

Jan 1, 1970
0
Jan Panteltje expounded in
...
I did one in the eighties with 8 x 8 or something like that,
controlled by a ZX81 Sinclair computer.

I finally got rid of a pair of those I had in my basement a couple of
years ago on ebay. Not sure what the buyer planned to use them for.
Also I think I used opamps in the drive lines and read lines.

I'm not sweating those details at this point in time. The idea is to be
able to replace the rats nest of patch cords, and make them restorable.
But I do not remember all the details apart from that it had a lot of
logic level shifters to drive those chips from 5V logic.

The newer chips do all that level shifting for you.
These day may be digitise everything do it in software?

This is for an analog synth. If I was to digitize the signal, then I
might as well forget about the analog synth too - where is the fun in
that? :)
I remember the ZX81 crashed one evening and caused a hell of a lot of
noise in the auditorium where it was used..
The program was written in BASIC.
Also reloading the program from tape took some time... before it was
online again LOL. But is was sort of a first :)

The British group Ultravox used to get really annoyed with the joint's
tech people who would turn off and then on the breakers to show them
where the breakers were. This of course forced them to have to reload all
of their digital effects from tape again.

Ultravox btw, is finally coming out with a new album 24 years after their
last one. They've all gone bald too <snicker>.

Warren
 
W

Warren

Jan 1, 1970
0
Syd Rumpo expounded in
Don't know if it helps, but look at AD75019 - it's still
available and is a IIRC +/-12V 16x16 analog crosspoint.
Annoying part is serial addressing, but a slave PIC would
do this easily.

Yes, the +/-12V is great, & matrix size too. The serial
addressing is ok with me for simpler AVR wiring. This part
has 200 ohm "on" resistance, which is a bit higher than the
typical 50, but it otherwise has the same issue. However, with
the larger matrix size it might work if I fan out the inputs
to multiple lines.

Warren
 
J

Joerg

Jan 1, 1970
0
Warren said:
Joerg expounded in
Warren said:
Joerg expounded in [Warren] wrote:
I'm trying to come up with a microprocessor solution to
replace the traditional analog synth patch cords, using an
analog crosspoint switch.

The idea is to replicate the mechanical matrix used in
this synth, but under microprocessor control: ..
Yet in the mechanical plug version, X3 and X4 have some level
of isolation (20k) to allow for different output signals. Y5
and Y6 also keep at least 20k worth of isolation from each
other. This is an oversimplification with a complex network,
but you get the idea hopefully. ..
But here, I'm looking for a way to route and mix, albeit with
limited isolation.
Ok ... but ... unless I am having a brain freeze here:

Suppose you send something from Y5 and Y6 into X4. Y5 also goes to
X3, which happens to also get a feed from some other place, and so
on. If you feed the mux strictly current mode and make sure you have
current receivers on all X ports then the impedance at all Y ports is
very high (ideally infinity). Yet at X3, X4 and the others it is
zero. Like a dead short. So how could there be any continuing leakage
for example from whatever else feeds into X3 via Y5 and then into X4?

Ok, using the Y5, Y6, X3 and X4 example, let's look at this from the
mechanical plug board pov first, that I am trying to replicate in
silicon. Remember each plug has a resistor in it for isolation.

Y5 is connected to X3 and X4
Y6 is connected to X4 only

The plugs are put in at the crosspoints, each of them containing a
resistor (I've assumed 10k ohms) :

(switch to fixed fonts here)

Y5 Y6
| |
| |
--|-----+--|-------- X3
+R=10k+ |
| |
| |
--|-----+--|-----+-- X4
+R=10k+ +R=10k+
| |
| |

This small matrix patch configuration is equivalent to:

Y5--+-R=10k--------X3
|
+-R=10k---+
|
Y6----R=10k---+----X4

From this, it is very easy to see that the isolation from X3 and X4 is
20k and also that Y5 and Y6 is also isolated by 20k. Y6 to X3 isolation
isn't great, but it is 30k ohms.

Yup, that's pretty much how I understood your explanation in the
previous post.

But when you use the silicon matrix solutions, the R is lowered to
roughly 50 ohms (or less). So then X3 and X4 is only isolated by 100 ohms
and likewise Y5 and Y6 is also only isolated by 100 ohms (like I said
earlier). This is completely unsuitable for voltage based signals (which
these are). Further Y6 is only 150 ohms away from X3.

What I think you're suggesting is to convert from a voltage based signal
and work with currents instead. The problem is that it must also accept
DC control signals (inputs are not always strictly ac signal). To apply
two conversions (voltage -> current and back) would introduce too much DC
error. Pitch control for example, is very fussy.

Depends on how much DC error can be tolerated. If you are talking low
microvolts, yeah, that'll get ugly. Then you can't use one of the
telco-style crosspoint ICs, that only works current mode. Problem is,
the old ones have mostly vanished. Even the ones from my days (22106 and
such) are gone.
 
Top