Differential ADC Inputs

 Hi Clyde, may be TI add to new devices, 10/12 bit AD converter are SAR  so 60dB and 72dB respectively  of dynamic range has no sense to make input differential.

 On 1.5V reference LSB value is near 0.366mV for 12Bit and 1.4mV for 10Bit.

Sigma delta are 16 or 24 bit resolution so this make sense to have differential inputs to cancel or at almost reduce noise. As from I write this there where 9 part 16bit and 48 part 24Bit Sigma delta converter.

 This case with 96.3 and 144.48dB (theoretical) dynamic range noise impact many bit. LSB value for a 1.5V ref is 22.88uV on 16 bit and 89nV(theoretical) on 24bit

 Many device has some sort of programmable OPAMP. Check family selector to see specific device.

 Regards

 Roberto

Resolution is irrelevant if the ground plane is noisy.  Differential inputs eliminate this problem.  It is far superior. 

Clyde Eisenbeis said:

Resolution is irrelevant if the ground plane is noisy.

Addign a differential stage to the inpu introduces additional linearity and offset problems. Which are difficult to resolve on a single supply. Even on the MSPs with differential input, this is NOT suitable to eliminate GND noise by a poor layout. It is to eliminate offset voltages and expand the ADCs resolution to the area of interest. E.g. if the input signal is between 2.0 and 3.0V, a differential input allows you using the more precise and higher resolution 1.5V reference instead of VCC. But since no MSP has a symmetrical supply, GND noise removal cannot be done with the differential input. Except if it is really heavy (and then you'll have otehr, bigger problems)

Jens-Michael Gross said:

Resolution is irrelevant if the ground plane is noisy.

That's why many MSPs with ADC12 have separate AVCC and AVSS pins. To provide a non-noisy supply.[/quote]

 Hi Jens, not to speak about induced noise on poor designed differential stages, if wrong design induce some noise on SAR single ended version then is completely unusable on differential lines. Differential lines apply to low speed converter so bandwidth is limited and no superiority can remove the error. On a 12BIT 200KSAS design an opamp to remove noise must have a noise floor and a bandwidth too far from scope of MSP.

 So IMHO TI never produce a differential MSP with SAR. Is the ability of designer to remove noise from his design, if differential noise is your problem and noise is in the range of computing power a filtering must apply to remove it.

 I designed filters with a higher bandwidth on MSP so it is easy to filter out noise in the input signal otherwise it is not noise but a poor design.

 Regards

 Roberto

There is always some noise on the ground plane, regardless of the quality of the design ... multiple ground planes ... filtering.  Switching to a differential sigma-delta ADC can significantly improve the results. 

The difference was tested in real-life noisy environments ... a single ended SAR (12 bit) was getting around 12 bits of noise (12 / 2^12 = 0.293%).  The differential sigma-delta (16) was getting around 7 bits of noise (7 / 2^16 = 0.011%).

In addition, when working with sensors, the output is often differential ... either leg can be a higher voltage than the other.  Converting to single ended requires an instrumentation amp plus a bias voltage. Differential inputs eliminates this ... reduces cost and reduces real estate. 

Clyde Eisenbeis said:

There is always some noise on the ground plane, regardless of the quality of the design ... multiple ground planes ... filtering.

Clyde Eisenbeis said:

Switching to a differential sigma-delta ADC can significantly improve the results. 

Clyde Eisenbeis said:

a single ended SAR (12 bit) was getting around 12 bits of noise (12 / 2^12 = 0.293%).  The differential sigma-delta (16) was getting around 7 bits of noise (7 / 2^16 = 0.011%).

Clyde Eisenbeis said:

In addition, when working with sensors, the output is often differential

However, every child can cry for more. It's the job of an engineer to make the best from what is available. :)

1) and 2) The key is noise reduction, not bits of resolution. Differential provides noise cancellation.

3) Yes, averaging can help.  Using an RC filter with a break point of 1 Hz would work too ... which makes it worthless when faster response time is required.

4) Instrumentation amps are not required.  Amplifying differential in / differential out works well.

5) The last paragraph is quite insulting. 

Clyde Eisenbeis said:

1) and 2) The key is noise reduction, not bits of resolution. Differential provides noise cancellation.

 Yes and No but in case of design noise differential stage is far  more inaccurate: ground noise is from HF noise so averaging can be different and some dc induced by rectification than amplification produce more offset than cancellation.

 So what noise you need to cancel? What is your bandwidth? What level of signal are you dealing with? Are differential delta-sigma matching your design? what resolution do you need? What sampling rate?

 Without these fundamental design point all word are just free to air.

Clyde Eisenbeis said:

3) Yes, averaging can help.  Using an RC filter with a break point of 1 Hz would work too ... which makes it worthless when faster response time is required

 If sensor are very low bandwidth then no need to use differential to remove noise, just a good ground plane and power decoupling design need to.

Clyde Eisenbeis said:

4) Instrumentation amps are not required.  Amplifying differential in / differential out works well.

Which sensor are you speaking about?

Clyde Eisenbeis said:

5) The last paragraph is quite insulting.

 Again this wording has some sense in an unspecified field where only you know the sensor and what you need to do.. IMHO this appear to me as an example of how the post can appear to other than you. Sorry but I don't see insult but a suggestion.

SO be more clean and specific to gain a close answer to your problem otherwise these word can be appropriate too also if you feel as a bad example.

 Regards

 Roberto

1) Lower than 100 Hz.  10 bits of resolution is adequate.

3) The sensor is remote ... there is no ground plane between the sensor and the assy.  Some RC filtering can be incorporated on both ends using a differential in/out configuration.

4) Hall sensor.

Clyde Eisenbeis said:

1) Lower than 100 Hz.  10 bits of resolution is adequate.

 Hi Clyde, sorry for long delay before answering again, I got very busy and offline too.

 Lower than 100Hz signify just induction from power line can affect them, good wiring to sensor can completely remove this noise.

 10 bit of resolution signify (10*6.02)dB, a so small dynamic if not far from processor don't need special filtering nor differential noise cancellation, an RC on output of Opamp near sensor and an input LC-RC coupled with shielded twisted pair completely clean all noise coming from external sources.

 IMHO solution to question 3 is to mount one processor close to sensor or select a Digital output hall sensor.

 From what i read from your writing I don't see a solution nor the need for a so complex design, I feel a sigma-delta msp430f20x3 can fit your need without external opamps doing some digital filtering to your signal too at much lower cost.

 Regards

 Roberto

Clyde Eisenbeis said:

4) Hall sensor.

 Hi Clyde, in addition to preceding answer, I was thinking about Hall sensor and differential inducted noise, Hall sensor can catch all EMF noise in the 100Hz range so inducted noise is still part of sensor measurement, are you sure noise come from lines and not as measured magnetic field noise? Hall sensor @ near 100Hz is more and more sensitive than a transmission line too reading noise as well as signal. Transmission lines are differential in nature so EMF is in phase on both ground and signal lines then ground plane is of no help, an Ultraperm shield can be of help near sensor field sensitive area here but we have no idea of what are you doing and what noise type are affecting your measurement too.

 Best regards

 Roberto

The simplest solution is for TI to provide differential inputs for all ADC inputs, not just sigma-delta ADC's.   This is the least expensive, least real estate, and better performance.

460822 said:

The simplest solution is for TI to provide differential inputs for all ADC inputs, not just sigma-delta ADC's.   This is the least expensive, least real estate, and better performance.

Do you kow how Delta-Sigma and SAR ADCs work?

A differential input for an SAR means an additional, ultra-fast OpAmp high-precision and lots of additional internal analog routing. It might be interesting for you, but it is neither least expensive nor least real estate and for sure not better performance. But for sure lots of additional current consumption.

Roberto Romano said:

89nV(theoretical) on 24bit

Jason Work said:

Some background material on the subject of ΔΣ analog to digital converters

Jens, I'm curious about your post "nor least real estate and for sure not better performance".  Suppose you have a sensor located ten feet from your assembly ... four wire cable connecting the assy to the remote sensor (two for power / gnd, two for signals) ... the sensor has a differential output ... the assy has an MSP single input SARS ADC  ... enough real estate  to add parts to the sensor PCB, but not enough real estate to add any more parts to the assy.  What configuration would you use?

By "real estate" I believe Jens was referring to silicon real estate, i.e. the size of the die.  Speaking from the TI perspective we must consider the needs of all of our customers and applications.  Everything is a trade-off.  In this case we can potentially add differential A/D converters which might be exactly the thing needed in your application.  However, that adds size/cost to the actual silicon.  Now all of the other customers that don't have this requirement end up paying more money for the processor for a feature they don't need.

It's always very difficult to come up with a good chip specification.  Lots of wrangling happens over which peripherals are included, how many of each peripheral, how much flash/RAM, which peripherals are pin-muxed, etc.  It's a huge balancing act where we try to accommodate as many needs/wishes as possible, though ultimately a few things may need to be supplemented externally or you may need to choose a different device to get everything to fit all your needs.

Thanks for the prompt reply.  Differential ADC's have been needed by our company for many years ... we end up using a delta-sigma (end up using two MSP's) or compromising with one MSP.

No one in our group has taken the time to post this question.  There are, quite likely, others / companies who would like SAR ADC's to have differential input capability.

Regarding "wrangling", the pin out of the MSP would be identical with differential vs. single ended ADC's.  And folks implementing differential, when needed, would end up saving real estate, which is often precious ... and improved performance in noisy environments (not dependent on super quiet ground planes). 

460822 said:

... would like SAR ADC's to have differential input capability.

"The ADC12_B module supports fast 12-bit analog-to-digital conversions with differential and single-ended inputs"

MSP430FR59xx, MSP430FR58xx Mixed Signal Microcontroller

New info! ... missed  seeing that.  Unfortunately, the FRAM is not large enough, but this is a step in the right direction. 

Jason Work said:

"The ADC12_B module supports fast 12-bit analog-to-digital conversions with differential and single-ended inputs"

After all, it is not a real differential mode, as in feeding an OpAmp, but rather doing two subsequent conversions and presenting the difference in the ADC12MEMx register. Somwthing one could have done with the normal ADC12 too with a simple subtraction of the two results. "differential" still doesn't mean that any of the inputs may be negative, as it is allowed on the Delta-Sigmas with input OpAmp stage.

(and yes, with 'real estate' I meant 'on the die', which directly affects the production costs and the current consumption) . What someone does on his PCB is his own thing. And if the PCB is too small for the MSP and external needed hardware, then it could well be too small for the then bigger MSP (more pins) with all needed hardware internally.)

460822 said:

New info! ... missed  seeing that.  Unfortunately, the FRAM is not large enough, but this is a step in the right direction. 

In MSP430Ware, you can see that more FRAM parts are in development.  Stay tuned.