Small Signal Analog Question

[Noway2]

Part of a project that I am working on requires measurement of some low voltage analog signals. These signals will effectively be generated one of the two boards in the system. Ultimately, these signals need to be read by an ADC. The main logic board in the system has a decent 12 bit ADC on it, but the signals of interest are aquired on the other board, which is basically an IO board. The signals are on the order of 5 to 500mv (before any amplification).

I am considering two approaches to the problem of getting the signal from one board to the other. The first approach I am considering is using a second processor, such as a Microchip 16F87x with a 10 bit ADC on the IO board, performing the conversion there, and transmitting the information via an SPI link to the main logic board.

The other alternative I am considering would be to bring the analog signal up to the main logic board and utilize the ADC on that board. For cost and simplicity reasons, this is approach I am leaning towards. I am thinking that in order to do this, it will be necessary to utilize a differential tranmitter - receiver, possibly with twisted pair wire.

I would like to keep the number of distint connections between the boards to a minimum especially limiting the number of discrete wires. Ideally, the connection between the boards would consist of one or two ribbon cables, which will also carry digital signals.

No matter how I look at the problem, I am seeing that it is a compromise between cost, complexity, and signal quality/ Does anybody have any suggestions to handle this situation?

 

[Madcow]

How many analog signals?
How many bits of resolution do you want?
How fast sampling?
How long is the cable?

The micro or a multi channel adc maybe better from a
noise point of view. I don't like mixing analog and digital
on the same ribbon cable.
When measuring a 0-50 mv shunt I also find a common mode choke/filter on the input helps the measurement.
Also after about 12 bits of resolution layout becomes very important.

 

[Noway2]

Madcow,

Thank you for your reply.

There will upto 8 analog channels. The one board in the system utilizes a DSP with a 12 bit ADC and (in my opinion) is capable of a ridicuously fast sample rate. The application, however, is VERY slow with regards to its speed requirements. The cable will probably be about 3 feet based on the present product that is being replaced. All of the analog signals of interest will be DC or slowly varying DC. If I decide to utilize the second processor, it will likely be a 10 bit with a 5V reference, which I think is a little bit anemic for measuring signals as low as .005V unless an gain amplifier stage is used.

My thoughts are leaning towards the concept of using a flat ribbon cable (seperate from the digital cable) and using a differential transmitter and reciever with a "S+ GND S- GND
" configuration in the cable with the grounds connecting to analog common, and placing the analog signals in one cable.

Could you please elaborate a little bit on the common mode choke / filter you mention. I am almost grasping what you are swaying, but I am having trouble visualizing the circuit implementation.

 

[IRstuff]

If it's that slow you could mux the 8 channels onto a single line.

TTFN

 

[zappedagain]

Noise is the big issue here. If you can afford the ADC on the IO board that is the cleanest with all digital signals on the cable (as long as your sampling circuit doesn't introduce a bunch of noise on the IO board.

For your differential signal cable GND S+ S- GND is a better layout to minimize the area between the differential lines; your layout was optimum for single ended signals.

One nice thing about "a rediculously fast sample rate" is that you can apply digital filtering. I.e. take 100 samplea and average them. If your noise is random (not systematic) you'll improve your signal to noise ratio by 10 (sqrt of average factor).

Madcow has some good questions that you didn't answer yet - what resolution and how often you need to sample (define "slow" - once a second? Once a day)? Those can both throw a lot of variation into the solution.

Check out

http://www.steward.com

for info on common mode filters.

 

[Noway2]

The idea of keeping the converter as close to the signal source as possible seems to be the consensus. I am juggling different ways to accomplish this goal, all of which have benefits and undesirables associated with them.

The answer to the questions of how fast and at what resolution, are quite honestly something that I have been asking myself. The marketing department here has been unable to provide any guidance as to what the customer needs, so I, the design engineer, have decided upon the following:

The range shall be 0 - 30Vdc and I would like a displayable resolution of .1 volts. A 10 bit converter should be able to provide, this to my satisfaction, assuming I get a good 9 bits out of it. As far as speed goes, the output shall be updated every .5 seconds. The 12 bit converter I have on the main processor has a conversion time of 80ns. From what I have determined from the datasheet parameters, it should be more than sufficient in its performance and accuracy. I plan to take advantage of the fact that I can sample at a faster rate and implement a digital filter, reducing the anti-aliasing filter requirements.

The problem looks like it is coming down to is it better to bring the analog signals to the main logic board or put a converter or processor on the I/O board. The first approach minimizes hardware cost, the second approach potentially saves some discrete wiring (a week point in our labor force) and standardizes the controller for what I plan for the 2nd phase of the project where it will be used in a very similar product but with different IO requirements.

 

[logbook]

Don't expect 9 "good bits" from a 10 bit converter. Cheap CMOS ones can be worse than that.

Systematic noise can be reduced faster than random noise when averaged. You just have to average over the right number! Averaging over an integer number of mains cycles is a good trick for removing mains hum for example. (This is how 6 1/2 digit DMMs get such quiet readings.)

 

[Noway2]

As much as one could hope for it, my experience says that you are right that it shouldn't be expected to get 9 "good bits" out of a 10 bit ADC. The product that I am ultimately going to replace with this new one uses a 10 bit ADC and I am lucky to get 8 bits out of it on a clear day. I think that this, coupled with a bunch of other layout and poor noise situation choices goes along way to explain why the circuit doesn't work too well.

 

[Warpspeed]

Well, I take it that accuracy and noise, and minimum number interconnections are prime requirements, and speed is not.

As already suggested I would multiplex the analog signals at the source and use something like a 12 bit multi slope integrating a/d converter. These are low cost and certainly have the required accuracy. The twelve parallel bits could always be converted to a serial bit stream if there is an unused serial port available on the main processor.

 

[logbook]

You can buy 8 channel 12 bit serial output ADCs so you really don't have much analog design work to do at all (which is the ideal situation!). A quick search on the Analog Devices site showed 11 ADCs with SPI interfaces so the total number of devices to choose from is rather high.

For a "DC" application like this there is no need to try very hard

 

[Noway2]

No need to try very hard??? Now you tell me after I just finished pulling the rest of my hair out . Oh well, I guess I didn't have that much left anyway.

Actually, using an SPI interfaced isn't a bad idea. If I put the ADC on the back panel, I will probably use an SPI interface as I had originally planned on using this technique to communicate with the panel card(s).

I was discussing this issue with my boss yesterday and I was mentioning that there are a couple of different ways that we can go (as to where to put the ADC), each of which has advantages and disadvantages.

I got tired of going in circles on this issue so I put it aside for now and started concentrating on a prototype of the analog circuitry itself.