Forums

pretty stange board

Started by John Larkin September 29, 2015

We make a lot of weird products, in VME and little Ethernet boxes.
They use D25, D37, SCSI50, SCSI68, and D9 i/o connectors. Some
products follow sorta repeatable channel/pin assignments, and some
don't.

We now have a motley collection of test sets and fixtures, some
historical creaky DOS/ISA/Bit3 stuff. The new regime will be one
universal test box, and Ethernet connection to the test gear, code in
Python.

We'll use a Fluke benchtop DVM (the Keithleys suck) and an Agilent SMU
(the Keithleys suck) and a rackmount scope, probably a Rigol. Maybe an
ARB, again a Rigol.

My assignment is to design the main board, which is  basically a relay
matrix. It goes something like this:

https://dl.dropboxusercontent.com/u/53724080/PCBs/P950_Block_1.jpg

There are 68 "PinN" signals, all used on a SCSI-68 connector and less
used on others. Each pin has three DPDT relays so it can be left open
or connected to one of five internal busses, A B C D or E. Some of the
pins can also be locally grounded; there are reasons to do that, trust
me.

https://dl.dropboxusercontent.com/u/53724080/PCBs/P950_Pin_Matrix_1.jpg

Then there is a "test matrix" that can ground the busses, short them,
and connect them to the test equipment. More relays. There are also
eight resistors that can be connected to A B C D in 4-wire mode, or
fake 4-wire mode for the higher values. They simulate RTDs and such.

https://dl.dropboxusercontent.com/u/53724080/PCBs/P950_Resistors_1.jpg

An ARM cpu manages it all, actually pretty simple since it mostly
clicks relays. It's looking like I'll have about 260 relays on the
board.

We'll use the TI TPIC6595 SPI relay drivers.... 32 or so of them. I
figure that if I allow the Fluke to monitor the +5 supply current, we
can check the coil currents and verify that the shift register and
coils are OK. Actual verification of the entire test set will require
some loopback connectors, which can be little paddle boards.

A test stand will need two of these boards, since many of our products
have two i/o connectors.

We can build a dozen of these boards, and keep spares in the closet.
Three reels of relays ought to do it.

I wish I could buy a universal crossbar switch!


On 29/09/2015 1:09 PM, John Larkin wrote:
> > > We make a lot of weird products, in VME and little Ethernet boxes. > They use D25, D37, SCSI50, SCSI68, and D9 i/o connectors. Some > products follow sorta repeatable channel/pin assignments, and some > don't. > > We now have a motley collection of test sets and fixtures, some > historical creaky DOS/ISA/Bit3 stuff. The new regime will be one > universal test box, and Ethernet connection to the test gear, code in > Python. > > We'll use a Fluke benchtop DVM (the Keithleys suck) and an Agilent SMU > (the Keithleys suck) and a rackmount scope, probably a Rigol. Maybe an > ARB, again a Rigol. > > My assignment is to design the main board, which is basically a relay > matrix. It goes something like this: > > https://dl.dropboxusercontent.com/u/53724080/PCBs/P950_Block_1.jpg > > There are 68 "PinN" signals, all used on a SCSI-68 connector and less > used on others. Each pin has three DPDT relays so it can be left open > or connected to one of five internal busses, A B C D or E. Some of the > pins can also be locally grounded; there are reasons to do that, trust > me. > > https://dl.dropboxusercontent.com/u/53724080/PCBs/P950_Pin_Matrix_1.jpg > > Then there is a "test matrix" that can ground the busses, short them, > and connect them to the test equipment. More relays. There are also > eight resistors that can be connected to A B C D in 4-wire mode, or > fake 4-wire mode for the higher values. They simulate RTDs and such. > > https://dl.dropboxusercontent.com/u/53724080/PCBs/P950_Resistors_1.jpg > > An ARM cpu manages it all, actually pretty simple since it mostly > clicks relays. It's looking like I'll have about 260 relays on the > board. > > We'll use the TI TPIC6595 SPI relay drivers.... 32 or so of them. I > figure that if I allow the Fluke to monitor the +5 supply current, we > can check the coil currents and verify that the shift register and > coils are OK. Actual verification of the entire test set will require > some loopback connectors, which can be little paddle boards. > > A test stand will need two of these boards, since many of our products > have two i/o connectors. > > We can build a dozen of these boards, and keep spares in the closet. > Three reels of relays ought to do it. > > I wish I could buy a universal crossbar switch! > >
A long time ago back in analog history there used to be arrays of crossbar switches, something of the order of 100 * 100 lines. Havent seen anything like that in years. I presume phone line and similar switching is now all done with CMOS analog switches, ruggedised versions of the venerable 4016 etc. Good luck.
On Tue, 29 Sep 2015 15:33:35 +1000, Adrian Jansen <adrian@qq.vv.net>
wrote:

>On 29/09/2015 1:09 PM, John Larkin wrote: >> >> >> We make a lot of weird products, in VME and little Ethernet boxes. >> They use D25, D37, SCSI50, SCSI68, and D9 i/o connectors. Some >> products follow sorta repeatable channel/pin assignments, and some >> don't. >> >> We now have a motley collection of test sets and fixtures, some >> historical creaky DOS/ISA/Bit3 stuff. The new regime will be one >> universal test box, and Ethernet connection to the test gear, code in >> Python. >> >> We'll use a Fluke benchtop DVM (the Keithleys suck) and an Agilent SMU >> (the Keithleys suck) and a rackmount scope, probably a Rigol. Maybe an >> ARB, again a Rigol. >> >> My assignment is to design the main board, which is basically a relay >> matrix. It goes something like this: >> >> https://dl.dropboxusercontent.com/u/53724080/PCBs/P950_Block_1.jpg >> >> There are 68 "PinN" signals, all used on a SCSI-68 connector and less >> used on others. Each pin has three DPDT relays so it can be left open >> or connected to one of five internal busses, A B C D or E. Some of the >> pins can also be locally grounded; there are reasons to do that, trust >> me. >> >> https://dl.dropboxusercontent.com/u/53724080/PCBs/P950_Pin_Matrix_1.jpg >> >> Then there is a "test matrix" that can ground the busses, short them, >> and connect them to the test equipment. More relays. There are also >> eight resistors that can be connected to A B C D in 4-wire mode, or >> fake 4-wire mode for the higher values. They simulate RTDs and such. >> >> https://dl.dropboxusercontent.com/u/53724080/PCBs/P950_Resistors_1.jpg >> >> An ARM cpu manages it all, actually pretty simple since it mostly >> clicks relays. It's looking like I'll have about 260 relays on the >> board. >> >> We'll use the TI TPIC6595 SPI relay drivers.... 32 or so of them. I >> figure that if I allow the Fluke to monitor the +5 supply current, we >> can check the coil currents and verify that the shift register and >> coils are OK. Actual verification of the entire test set will require >> some loopback connectors, which can be little paddle boards. >> >> A test stand will need two of these boards, since many of our products >> have two i/o connectors. >> >> We can build a dozen of these boards, and keep spares in the closet. >> Three reels of relays ought to do it. >> >> I wish I could buy a universal crossbar switch! >> >> > >A long time ago back in analog history there used to be arrays of >crossbar switches, something of the order of 100 * 100 lines. Havent >seen anything like that in years. I presume phone line and similar >switching is now all done with CMOS analog switches, ruggedised versions >of the venerable 4016 etc. > >Good luck.
Phones are all digital. Each POTS phone line gets an ADC, a DAC, and some sort of DC feed/ringer. No more Strowger relays or crossbars. I guess the exchanges are silent now. They used to sound like a machine-gun nest. My exchange doesn't seem to accept pulse dialing any more.
On 9/29/2015 1:33 AM, Adrian Jansen wrote:
> On 29/09/2015 1:09 PM, John Larkin wrote: >> >> >> We make a lot of weird products, in VME and little Ethernet boxes. >> They use D25, D37, SCSI50, SCSI68, and D9 i/o connectors. Some >> products follow sorta repeatable channel/pin assignments, and some >> don't. >> >> We now have a motley collection of test sets and fixtures, some >> historical creaky DOS/ISA/Bit3 stuff. The new regime will be one >> universal test box, and Ethernet connection to the test gear, code in >> Python. >> >> We'll use a Fluke benchtop DVM (the Keithleys suck) and an Agilent SMU >> (the Keithleys suck) and a rackmount scope, probably a Rigol. Maybe an >> ARB, again a Rigol. >> >> My assignment is to design the main board, which is basically a relay >> matrix. It goes something like this: >> >> https://dl.dropboxusercontent.com/u/53724080/PCBs/P950_Block_1.jpg >> >> There are 68 "PinN" signals, all used on a SCSI-68 connector and less >> used on others. Each pin has three DPDT relays so it can be left open >> or connected to one of five internal busses, A B C D or E. Some of the >> pins can also be locally grounded; there are reasons to do that, trust >> me. >> >> https://dl.dropboxusercontent.com/u/53724080/PCBs/P950_Pin_Matrix_1.jpg >> >> Then there is a "test matrix" that can ground the busses, short them, >> and connect them to the test equipment. More relays. There are also >> eight resistors that can be connected to A B C D in 4-wire mode, or >> fake 4-wire mode for the higher values. They simulate RTDs and such. >> >> https://dl.dropboxusercontent.com/u/53724080/PCBs/P950_Resistors_1.jpg >> >> An ARM cpu manages it all, actually pretty simple since it mostly >> clicks relays. It's looking like I'll have about 260 relays on the >> board. >> >> We'll use the TI TPIC6595 SPI relay drivers.... 32 or so of them. I >> figure that if I allow the Fluke to monitor the +5 supply current, we >> can check the coil currents and verify that the shift register and >> coils are OK. Actual verification of the entire test set will require >> some loopback connectors, which can be little paddle boards. >> >> A test stand will need two of these boards, since many of our products >> have two i/o connectors. >> >> We can build a dozen of these boards, and keep spares in the closet. >> Three reels of relays ought to do it. >> >> I wish I could buy a universal crossbar switch! >> >> > > A long time ago back in analog history there used to be arrays of > crossbar switches, something of the order of 100 * 100 lines. Havent > seen anything like that in years. I presume phone line and similar > switching is now all done with CMOS analog switches, ruggedised versions > of the venerable 4016 etc.
I remember the crossbar race, chip makers trying to get more and more switches on a single chip. That was about the same time telephony went digital which means it was all handled very differently with digital muxes rather than analog switches. The crossbar race ended very quickly, almost as soon as it started. Digital is so fast, switching can be done in time much, much more efficiently than analog could ever hope to achieve. That said, telephony is pretty impressive stuff. I have been told they have such stringent requirements for lost bits and dropped calls that even with digital stuff it can be hard to meet the requirements. If you drop 1 in 10,000 calls, how do you troubleshoot that? -- Rick
On Tuesday, 29 September 2015 04:10:08 UTC+1, John Larkin  wrote:
> We make a lot of weird products, in VME and little Ethernet boxes. > They use D25, D37, SCSI50, SCSI68, and D9 i/o connectors. Some > products follow sorta repeatable channel/pin assignments, and some > don't. > > We now have a motley collection of test sets and fixtures, some > historical creaky DOS/ISA/Bit3 stuff. The new regime will be one > universal test box, and Ethernet connection to the test gear, code in > Python. > > We'll use a Fluke benchtop DVM (the Keithleys suck) and an Agilent SMU > (the Keithleys suck) and a rackmount scope, probably a Rigol. Maybe an > ARB, again a Rigol. > > My assignment is to design the main board, which is basically a relay > matrix. It goes something like this: > > https://dl.dropboxusercontent.com/u/53724080/PCBs/P950_Block_1.jpg > > There are 68 "PinN" signals, all used on a SCSI-68 connector and less > used on others. Each pin has three DPDT relays so it can be left open > or connected to one of five internal busses, A B C D or E. Some of the > pins can also be locally grounded; there are reasons to do that, trust > me. > > https://dl.dropboxusercontent.com/u/53724080/PCBs/P950_Pin_Matrix_1.jpg > > Then there is a "test matrix" that can ground the busses, short them, > and connect them to the test equipment. More relays. There are also > eight resistors that can be connected to A B C D in 4-wire mode, or > fake 4-wire mode for the higher values. They simulate RTDs and such. > > https://dl.dropboxusercontent.com/u/53724080/PCBs/P950_Resistors_1.jpg > > An ARM cpu manages it all, actually pretty simple since it mostly > clicks relays. It's looking like I'll have about 260 relays on the > board. > > We'll use the TI TPIC6595 SPI relay drivers.... 32 or so of them. I > figure that if I allow the Fluke to monitor the +5 supply current, we > can check the coil currents and verify that the shift register and > coils are OK. Actual verification of the entire test set will require > some loopback connectors, which can be little paddle boards. > > A test stand will need two of these boards, since many of our products > have two i/o connectors. > > We can build a dozen of these boards, and keep spares in the closet. > Three reels of relays ought to do it. > > I wish I could buy a universal crossbar switch!
What's your relay MTBF going to be? Do you use a routine of switch, self check, test UUT, self test? Or all SS relays? NT
John Larkin wrote:


> I wish I could buy a universal crossbar switch!
We built several of these for remote selection of signals from areas that are inaccessibe due to radiation. It uses two 16 x 8 crossbar arrays and 8 channels of buffer amplifiers. A Beagle Board computer controls it as an Ethernet appliance. So, you have 32 50-Ohm inputs and 8 outputs. Each output can select from any of the 32 inputs. Very handy! Jon
John Larkin wrote:


> I guess the exchanges are silent now. They used to sound like a > machine-gun nest.
#5 ESS used a huge array of ferreed switches, basically a giant core memory with biased reed switches in each core. The array magnetizes the core to either aid or cancel the bias magnet. it made a fascinating tinkling sound when you stood in the ferreed array, thousands of contacts a second making tiny plinks as the opened and closed. Not sure how many #5 ESS systems are still running, but I sort of think there are quite a few still running. Perhaps they have superseded the ferreeds with somthing totally electronic. Jon
Jon Elson wrote:


> > Not sure how many #5 ESS systems are still running, but I sort of think > there are quite a few still running. Perhaps they have superseded the > ferreeds with somthing totally electronic.
Well, a little wiki searching indicates that what I saw must have been a #1 ESS, not #5, with the ferreeds. Jon
On Tue, 29 Sep 2015 11:09:49 -0700 (PDT), tabbypurr@gmail.com wrote:

>On Tuesday, 29 September 2015 04:10:08 UTC+1, John Larkin wrote: >> We make a lot of weird products, in VME and little Ethernet boxes. >> They use D25, D37, SCSI50, SCSI68, and D9 i/o connectors. Some >> products follow sorta repeatable channel/pin assignments, and some >> don't. >> >> We now have a motley collection of test sets and fixtures, some >> historical creaky DOS/ISA/Bit3 stuff. The new regime will be one >> universal test box, and Ethernet connection to the test gear, code in >> Python. >> >> We'll use a Fluke benchtop DVM (the Keithleys suck) and an Agilent SMU >> (the Keithleys suck) and a rackmount scope, probably a Rigol. Maybe an >> ARB, again a Rigol. >> >> My assignment is to design the main board, which is basically a relay >> matrix. It goes something like this: >> >> https://dl.dropboxusercontent.com/u/53724080/PCBs/P950_Block_1.jpg >> >> There are 68 "PinN" signals, all used on a SCSI-68 connector and less >> used on others. Each pin has three DPDT relays so it can be left open >> or connected to one of five internal busses, A B C D or E. Some of the >> pins can also be locally grounded; there are reasons to do that, trust >> me. >> >> https://dl.dropboxusercontent.com/u/53724080/PCBs/P950_Pin_Matrix_1.jpg >> >> Then there is a "test matrix" that can ground the busses, short them, >> and connect them to the test equipment. More relays. There are also >> eight resistors that can be connected to A B C D in 4-wire mode, or >> fake 4-wire mode for the higher values. They simulate RTDs and such. >> >> https://dl.dropboxusercontent.com/u/53724080/PCBs/P950_Resistors_1.jpg >> >> An ARM cpu manages it all, actually pretty simple since it mostly >> clicks relays. It's looking like I'll have about 260 relays on the >> board. >> >> We'll use the TI TPIC6595 SPI relay drivers.... 32 or so of them. I >> figure that if I allow the Fluke to monitor the +5 supply current, we >> can check the coil currents and verify that the shift register and >> coils are OK. Actual verification of the entire test set will require >> some loopback connectors, which can be little paddle boards. >> >> A test stand will need two of these boards, since many of our products >> have two i/o connectors. >> >> We can build a dozen of these boards, and keep spares in the closet. >> Three reels of relays ought to do it. >> >> I wish I could buy a universal crossbar switch! > >What's your relay MTBF going to be? Do you use a routine of switch, self check, test UUT, self test? Or all SS relays? > > >NT
I'll use little surface-mount Fujitsu DPDT electromechanical relays. They have been very reliable. Of course, we'll have to test the test set regularly, once a week maybe. It's probably that a relay failure would fail a DUT test, so we'd catch most of those. It's really hard to beat a relay for on resistance, isolation, and electrical ruggedness.
Den tirsdag den 29. september 2015 kl. 05.10.08 UTC+2 skrev John Larkin:
> We make a lot of weird products, in VME and little Ethernet boxes. > They use D25, D37, SCSI50, SCSI68, and D9 i/o connectors. Some > products follow sorta repeatable channel/pin assignments, and some > don't. > > We now have a motley collection of test sets and fixtures, some > historical creaky DOS/ISA/Bit3 stuff. The new regime will be one > universal test box, and Ethernet connection to the test gear, code in > Python. > > We'll use a Fluke benchtop DVM (the Keithleys suck) and an Agilent SMU > (the Keithleys suck) and a rackmount scope, probably a Rigol. Maybe an > ARB, again a Rigol. > > My assignment is to design the main board, which is basically a relay > matrix. It goes something like this: > > https://dl.dropboxusercontent.com/u/53724080/PCBs/P950_Block_1.jpg > > There are 68 "PinN" signals, all used on a SCSI-68 connector and less > used on others. Each pin has three DPDT relays so it can be left open > or connected to one of five internal busses, A B C D or E. Some of the > pins can also be locally grounded; there are reasons to do that, trust > me. > > https://dl.dropboxusercontent.com/u/53724080/PCBs/P950_Pin_Matrix_1.jpg > > Then there is a "test matrix" that can ground the busses, short them, > and connect them to the test equipment. More relays. There are also > eight resistors that can be connected to A B C D in 4-wire mode, or > fake 4-wire mode for the higher values. They simulate RTDs and such. > > https://dl.dropboxusercontent.com/u/53724080/PCBs/P950_Resistors_1.jpg > > An ARM cpu manages it all, actually pretty simple since it mostly > clicks relays. It's looking like I'll have about 260 relays on the > board. > > We'll use the TI TPIC6595 SPI relay drivers.... 32 or so of them. I > figure that if I allow the Fluke to monitor the +5 supply current, we > can check the coil currents and verify that the shift register and > coils are OK. Actual verification of the entire test set will require > some loopback connectors, which can be little paddle boards. > > A test stand will need two of these boards, since many of our products > have two i/o connectors. > > We can build a dozen of these boards, and keep spares in the closet. > Three reels of relays ought to do it. > > I wish I could buy a universal crossbar switch!
Agilent 34972a with the right modules could probably do it 34904A, 4 x 8 Matrix Switch 34908A, 40 channel single ended multiplexer it measures voltage,current,resistance,frequency,period,rtc,thermistor, thermocouple and is easily controlled with SCPI on via serial port or tcpip -Lasse