Huge analog switching matrix

Question

My application is a PCB that has 2 different 100 pins connectors and has to be capable of arbitrarily connect each pin on the first connector to each pin of the second.

The signal going through the board, once ALL the connections are set, will be on only a subset (4) lines at any given time and will be in the DC-20 Hz band at a maximum of 1mA @ 0.1V. Cross-talk is not to be considered in this application.

I have to change the configuration once in a while and when I have to do that I can access the board manually. Active components (like MUXes) are discouraged because the signals are used for cryogenic 4-wires measures and the cost of suitable components quickly becomes inaccessible.

My original solutions have been:

• Using a set of DIP SPST switches to build the connection matrix
• Using a lot of small (16mm^2) push-down buttons to build the connection matrix

Both this solutions are expensive and space consuming. I was wondering if anyone knows a better component for this application.

The need for flexibility comes from the deployment environment of the application. Once (and if) this component is deployed in the field (South pole station) and set I won't have access to it for ~1y. So in 3 months I'll spend there I'll have to make sure everything works and there's the possibility I'll have to reroute some channels. The good old hydra cable soldering is not reliable, thus I was exploring other solutions.

Answer 1

With the freshly added info:

The need for flexibility comes from the deployment environment of the application. Once (and if) this component is deployed in the field (South pole station) and set I won't have access to it for ~1y. So in 3 months I'll spend there I'll have to make sure everything works and there's the possibility I'll have to reroute some channels. The good old hydra cable soldering is not reliable, thus I was exploring other solutions.

Well, this sounds like you should bring a soldering iron to the South Pole, some solder wick and solder:

Make a square board, with the input connector on the "north" edge of the board, from it extending 100 columns.

The output connector on the "west" edge, from it extending 100 rows.

Add 100 "solder jumpers", i.e. connections you can make with solder (and unmake with a wick), from each row (10,000 in total), and from the "free" end of these, place a via to the underlying column.

That way, you get a pretty inflexible (as in: soldering tools needed to reconfigure), but magnificently reliable connection matrix.

Answer 2

So, when you say

100 pins connectors and has to be capable of arbitrarily connect each pin on the first connector to each pin of the second.

and imagine one connector to be an input vector \$U\in \mathbb F^{100}\$, the other to be a result vector \$V \in \mathbb F^{100}\$, then your whole device would need to be a 100×100 permutation matrix \$\Pi\in \mathbb F^{100\times 100}\$. (i.e. a matrix where each row and column has a single unity element, the rest being zero):

$$V=\Pi U\tag1\label{full}$$

You can straightforward implement that using any kind of switches.

And such a matrix has 10,000 entries that can be 0 or 1 – i.e. 10⁴ switches; relays that you control from a PC would probably be wisest, simply to avoid human error.

Now, you say:

The signal going through the board, once ALL the connections are set, will be on only a subset (4) lines at any given time

Well, that makes it easier:

Instead of having \$\eqref{full}\$, you can imagine an intermediate result vector \$W\in \mathbb F^{4}\$.

If you don't even "blow this back up" to the full 100 outputs in \$V\$, and I don't see a reason for that, your measurement devices doesn't seem to care whether it scans 100 channels or just the 4, that boils down to

$$\tilde V = \mathbf PU,$$

with \$\mathbf P\in \mathbb F^{100\times 4}\$, i.e. only 400 switches instead of 10,000.

If you can further restrict that (for example, there's only specific index subsets from which the first of your four outputs can come), then you might further simplify the problem: For example, four 25×1 matrices are still way easier than one 100×4 one!

An order of 400 switches still isn't cheap, and now you'll have to switch dynamically when you want to measure a different 4-set of channels. However, signal relays (exactly for switched-circuit telephony historical reasons) are comparatively cheap (A list).

Considering coil voltages and currents of 5 V, 40 mA typ., you might just directly drive these off open-drain or open-collector shift registers for a couple of cents, e.g. STPIC6D595, which you can in turn directly attach to a 5V bitbang/SPI-driver (which exist either as microcontrollers or as USB-to-SPI converters, for example).

Rough calculation of the board you'll need: The relay I linked to has an area of 11×21 mm², and you need a 10×6 mm² shift register for every 8 relays.

That makes a total area of 400·231 mm² + 50·60 mm² = 95400 mm². If you're a bit symmetrical about arranging the 8 relays head-to-head and the shift register at the end of each such row, you end up with a ca 17 cm wide board; if you put 32 rows on a board, that will be, including auxillary things (connectors etc) be 40 cm tall. Manufactured in China, five boards of that size will set you back by ca 40 to 80€ incl. shipping.

Adding 440 relays and 55 shift registers (10% overorder in case you damage anything) for about 300 € (not including taxes); throw in 50€ for connectors, power supply, LEDs and stuff, this is a 400€ project.

Compare that to the 10,000 switches: rather cheap and easy!

(in case you're a PhD candidate physicist or similar: get a student research assistant to design, and assemble, and test, and write a minimal control software for that board. Pay her or him well, i.e. overprovision her hours if necessary. These kinds of devices, if done well, usually have long-term utility after the end of your PhD. Add a sticker to it that points to some public documentation, and put all design files under CERN's open hardware license v2, and put everything on the internet, e.g. github, so that the next physicist doesn't reinvent the wheel. People have been cited for less in other people's papers!)

Answer 3

The solution that I'd use as the prime candidate to compare others against is one which you mentioned in a comment

I'm looking into a matrix of pin headers and a bunch of jumpers to make the contacts. It's simpler than DIP switches or relays and cheaper.

While this is not as elegant as a somewhat more automated solution it may be as good or better in practice.

Given the large number of inputs and outputs, the infrequent requirement for change and the apparently non time critical nature of the connection alteration a jumper and header solution sounds very workable. Also cheap, easy and compact (choose any 3) compared to obvious alternatives.

These are available in various lengths and with either polarity plugs and sockets at each end. While the low cost Asian sourced ones seem to work commendably well in general use you'd need to obtain such from a source with known contact materials and supplier and manufacturer provenance. [Digikey is generally a good source of genuine product - but certainly not the only source.].

Long term connection reliability is obviously an issue.
Your signal specification of "DC-20 Hz band at a maximum of 1mA @ 0.1V." means "contact wetting" needs to be addressed. Many materials exhibit high or even extremely high contact resistance at low currents. You'll need to examine the properties of specific contact materials for suitability. This is an area that competent connector manufacturers are well versed in and technical help will be readily available. "Off the cuff" gold-gold should work well BUT do check this yourself.

__________________________________

Possibly tolerable alternatives are screw down connections where the screw tightening action provides low contact resistance, or DIN rail connectors with screw down connections. Costs and volume are not as low as for the jumper connections but still modest compared to the value of your project.

Here are links to a zillion DIN rail; connectors.

Here ius an Element 14 page for a Phoenix contacts DIN rail connector - $1.56 each in 250 quantity. Cheaper elsewhere probably.

3.2 mV drop at 24A if I read the datasheet correctly. (Caveat Emptor).

Datasheet here

____________________________________

If you REALLY wanted to save money you could have 100 + 100 wires with suitably terminated ends and 4 only connector blocks :-).

The above use screwdriver implemented termination.

You can get hold button - insert wire - release button versions which may be easier to use in polar conditions. Or lever release versions - image below. From here - NB example only. NOT a recommendation re brand.

Answer 4

I have built a board before that is similar to what you are describing.

Each pin on connector #1 connects to four solid state relays. Each of the four relays connects to one of four nets A, B, C, D. Thus any pin can be connected to either net A, B, C, D.

The same thing is done for connector #2.

Since you have four nets A, B, C, D you can connect any arbitrary combination of up to four pins on one connector, to any set of four pins on the other connector.

The total number of relays needed is 800. If its low current and low voltage, you may get 800 relays for under $1000. If its higher voltage or current then you may pay closer to $10,000.

To drive the relays you can use a latching shift register relay driver such as the TLC59281DBQR (which is under $1.00 each). You would need 50 of those to drive 800 relays.

The setup doesn't consume that much power since you only need 8 of the relay on at any one time. Depending on which relay drivers you pick they can consume micro-amps. And you can get a low power micro-controler consuming under 1 mA to power the whole thing.

If you need any test signals you can throw on a few more relays that connect test signals to any of the A, B, C, D nets.

If you really want to do it mechanically, you could do the exact same setup with buttons or toggle switches instead of relays.

Answer 5

An off the shelf Relay matrix board will solve your problem. Relays are passive, some of these boards use latching relays, so the routes maintain state even when the controller is not powered. Since you are switching signals and not power, a reed relay matrix is also a viable solution. You can get such boards from the usual suspects (Tek, Keysight, NI).