DIY high resolution multi-slope converter

[NNNI]

The thing that leaves me most confused about MS-IV is its apparent complexity, but at the same time upon analyzing the schematics it looks like a regular multislope with a couple of extras. The first one is the DAM switch, described in detail in US Patent 6876241. From what I can see at least, it's basically just a level shifter and a CMOS output stage, somewhat like a logic inverter. I assume they had to come up with this because analog switches were not fast enough for what they were planning. The second thing is that they read the integrator waveforms using a coarse ADC, and according to AoE 3 this is so they can keep the integrator bounded. A window comparator would have worked for that as well, but there must have been a good reason to do it using an ADC. I hope to figure most of this stuff out once I get a closer look at the ADC waveforms. I did some calculations about reading speed in this log: https://hackaday.io/project/190528-multislope-adc/log/219031-integrators-dam-switches-and-speed

Regarding multiple rundown slopes - do you think that might be one of the main reasons for low INL and noise? I assume that the ultimate limiting factor in a 34401A-style multislope is the residue ADC itself.

[Kleinstein]

The rundown part does not give good linearity. It is more like a way to get high resolution / low noise.
The MS3 ADC has one major weak point to start with: the limited resolution of the residual charge ADC. The resolution in (counts) is given by the auxiliary ADC resolution times the modulation frequency minus a little overhead.  Be sides the nominal resolution the ADC in the µC also has quite some noise (a few LSB). To get at least a little more resolution the modulation is chosen rather fast. By itself a good choice but it comes with compromises.

A second weak point comes from the relatively slow settling integrator, that needs quite some minimum length of the pulses and with the fast modulation this looses quite a bit of the input range. So quite a bit of the time is lost to the fixed ref. setting part. To still get a 12 V input full scale range this needed the effective divider with 100 K and 42 K to ground in the input path. The ADC would actually work better if the input would use 30 K and a 3 V input range instead - possibly an earlier design stage left over from the 3457. The divider at the input / only part of the range actually used amplifies quite a lot of the other noise sources in the ADC.

[NNNI]

Could you elaborate a little more about integrator settling time? I assume this would matter more during rundown or residue reading than during runup.

[Kleinstein]

After switching the reference there is a small voltage pulse on the integrator input and it takes some time for settling after that. For the simple 1 OP-amp integrator this is a step in the voltage so a more permanent voltage. With the 2 OP-amp integrator one has a pulse only. The settling time depends on the speed of the amplifiers and other circuit details (e.g. RC at the input, divider between the stages).
The rundown part has only a limited number of switching events, so the settling part can cause some error, but only a rather small one. Especially for the initial part one can make sure the pulses are not too short. The run-up part may be more relevant because of many switching events (e.g. some 1000), but the timing could be more predictable so that some constant effect can be seen as part of the offset

For the residual charge reading it helps to have a fixed time from the last switching event (for the measurement on the fly) or enough waiting time (for the residual charge after rundown).

[NNNI]

I see, and would adding a small value (100s of pF) capacitor to the summing junction help mitigate these effects? almost all multislope implementations I've seen have such a capacitor.

[Kleinstein]

Capacitance from the summing junction to ground may help with the very fast part, but in most cases for the main visible settling it does not help or even makes things ring more.
To effect the settling it is more an RC series element that can make a difference in dampening ringing.

[Alex Nikitin]

In my ADC I use the LV4053 with 5 V supply and a measured on resistance of some 20 Ohm. With 10 K resistors (and 10 V at the input) the INL contribution is quite significant already. So even the TMUX1134 may want a little more than 10 K - maybe 20 K as a resonable lower limit. Besides the R_on a higher supply voltage also help in reducing the voltage dependence.
The 50 K resistors already allow quite low a noise and noise wise there is little need to go much lower with a 10 V or similar input range.

Perhaps ADG6412 worth a try? 0.5 Ohm and exceptionally linear across the voltage range.

Cheers

Alex

[Kleinstein]

For my ADC configuration with switching at the integrator input there is no need for a 20 V switch. It is more about a good balance of on resistance, capacitance and charge injection.
Possible alternatives would be more other low voltage switches like max4619 or the aready mentioned TMUX1133.

The LV4053 with some 20 Ohm (5 V) is not such a bad choice and OK with 50 K input resistors. The nonlinear contribution is small (e.g. 0.1 ppm range) and if needed one could numerical correct for some of this or as in my current front end do an internal averaging over positive and negative reading, so that the U² part does not matter that much.
Even just 2 x LV4053 in parallel may work if less than 50 K for the input are wanted.

With smaller resistors also the self heating part gets more tricky and this part can be temperature dependent  and thus more difficult to compensate for. 

[dietert1]

Others use active current sources. The Advantest R6581T schematic shows active current sources for the unknown and for the principal rundown current. In the Prema 6048 the unknown is a current source, too. In some sense the rundown current source is active, too, as it includes a sense line to correct for the switch behaviour. Looks like active current sources are appropriate for a 8.5 meter.
To convert the run-down current source of a Prema 6048 into an active one for near ground current routing, one can use a dual opamp, a small mosfet and 2x 1K and 1x 10K resistors of an LT5400. The divider raises the reference from 7 to 7.64 V and then senses the current at 0.64 V. That is the headroom for the mosfet, the switch and maybe some kind of filter. One might use something similar for the unknown. A little more complicated if it needs to be bipolar.

Regards, Dieter

[Kleinstein]

Active current sources offer some advantage, but also make the circuit more complicated.
As a downside one needs more complicated reference voltage levels and tends to loose a bit headroom. On the upside the switches can be low capacitance as the resistance does not matter that much. In the 6581 circuit one sees quite some effort (JFET+2 OP-amps) needed to make the current sources fast settling.
For the input part a current source is tricky - the prema solution with the extra supply is nice, but prone to pic up hum from the supply. This may need the PLL that is otherwise not that attractive.
A bipolar active current source for the input is more than just a little more tricky. Limiting the input to 1 polarity is a bit limiting.

One could combine current sources for the reference with a very low resistance (e.g. 1 Ohm range) for the signal input. Matching in the resistance / switches is to get TC compensation for the combined switch + resistor combinations. With a very kow on resistance the compensation is no longer needed. The input switch is only switched on/off once per conversion and capacitance thus less relevant.

For higher input and reference curent (to get even lower noise) active current source are definitely an option.
My ADC circuit was initial designed to get a rather simple multi-slope ADC. The higher performance target was later with a few minor changes (change HC4053 to LV4053, added sync flip-flops and adding filtering for the reference).

[NNNI]

I was doing some LTSpice simulations of a composite integrator, and out of curiosity I tried different dividers on the output of the "slow" input stage op-amp. Looks like the divider ratio does determine how fast the integrator settles after the input current changes, just like Kleinstein said.