Simplest Analog to Digital Converter

[PauloConstantino]

http://tinyurl.com/y8mldodq

A 8 bit SAR ADC build in falstads simulator for you to have a play with.

Full link incase it breaks later on.

Code: [Select]

http://www.falstad.com/circuit/circuitjs.html?cct=$+1+0.000005+10.20027730826997+64+5+50%0Ar+-64+0+-64+-64+0+100000%0Ar+-16+0+-16+-64+0+100000%0Ar+32+0+32+-64+0+100000%0Ar+-64+-64+-16+-64+0+50000%0Ar+-16+-64+32+-64+0+50000%0Ar+-160+-64+-112+-64+0+100000%0Aw+432+-64+432+-16+0%0AO+432+-64+496+-64+1%0Ar+-64+-64+-112+-64+0+50000%0Ar+-112+-64+-112+0+0+100000%0Ar+32+-64+80+-64+0+50000%0Ar+80+-64+128+-64+0+50000%0Ar+128+-64+176+-64+0+50000%0Ar+80+0+80+-64+0+100000%0Ar+128+0+128+-64+0+100000%0Ar+176+0+176+-64+0+100000%0Ar+224+0+224+-64+0+100000%0Ar+176+-64+224+-64+0+50000%0Aw+320+-64+224+-64+0%0A164+240+64+384+64+1024+4+0+0+0+5+false%0A164+240+224+384+224+1024+4+0+5+5+0+false%0Aw+240+224+240+192+0%0Aw+240+192+368+192+0%0Aw+368+192+368+64+0%0Aw+368+64+336+64+0%0Aw+240+64+224+64+0%0Aw+224+64+224+0+0%0Aw+240+96+176+96+0%0Aw+176+96+176+0+0%0Aw+240+128+128+128+0%0Aw+128+128+128+0+0%0Aw+240+160+80+160+0%0Aw+80+160+80+0+0%0Aw+240+224+32+224+0%0Aw+32+224+32+0+0%0Aw+240+256+-16+256+0%0Aw+-16+256+-16+0+0%0Aw+-64+0+-64+288+0%0Aw+-64+288+240+288+0%0Aw+-112+0+-112+320+0%0Aw+-112+320+240+320+0%0Aa+640+0+720+0+1+5+0+1000000+4.98624995111725+0.4296874999999993%0AR+640+16+576+16+0+1+5+2.5+2.5+0+0.5%0Aw+640+-16+432+-16+0%0AM+720+0+784+0+0+2.5%0A168+-256+224+-192+224+1024+4%0A168+-256+64+-208+64+1024+4%0Aw+-112+320+-160+320+0%0Aw+-64+288+-160+288+0%0Aw+-16+256+-160+256+0%0Aw+32+224+-160+224+0%0Aw+80+160+-160+160+0%0Aw+128+128+-160+128+0%0Aw+176+96+-160+96+0%0Aw+224+64+-160+64+0%0Aw+-160+192+-128+192+0%0Aw+-128+192+-128+352+0%0Aw+-160+352+-128+352+0%0Aw+-128+352+720+352+0%0Aw+720+352+720+0+0%0AM+-256+64+-288+64+0+2.5%0AM+-256+96+-288+96+0+2.5%0AM+-256+128+-288+128+0+2.5%0AM+-256+160+-288+160+0+2.5%0AM+-256+224+-288+224+0+2.5%0AM+-256+256+-288+256+0+2.5%0AM+-256+288+-288+288+0+2.5%0AM+-256+320+-288+320+0+2.5%0AR+336+224+448+224+1+2+10000+2.5+2.5+0+0.5%0Aw+320+-64+432+-64+0%0Ag+-160+-64+-192+-64+0%0Ax+474+233+591+236+4+24+ADC%5CsClock%0Ao+42+64+0+4099+5+0.00009765625+0+2+42+3%0Ao+60+64+0+4098+5+0.1+1+1%0Ao+61+64+0+4098+5+0.1+1+1%0Ao+62+64+0+4098+5+0.1+1+1%0Ao+63+64+0+4098+5+0.1+1+1%0Ao+64+64+0+4098+5+0.1+1+1%0Ao+65+64+0+4098+5+0.1+1+1%0Ao+66+64+0+4098+5+0.1+1+1%0Ao+67+64+0+4098+5+0.1+1+1%0A

+1 this is very neat

Yay I thought I was the only one using falstad 

[Yansi]

I have a few power Wayne Kerr supplies whose front panel display board implements a ADC using one op-amp, one comparator, and one CMOS analog switch IC, plus a voltage reference. The basic 8051 CPU controls the analog switch and looks at the comparator output. Very simple. It achieves 12 bit resolution (>3500 counts) at maybe 10 samples per second. It's a basic dual slope design: switches either the voltage or current signal to integrate into a capacitor for a fixed period, then switches to the opposite polarity voltage reference and measures de-integration time.

I have offered a dual slope integrating ADC on the previous page already, but everyone seems to ignore it. It is I think the best bang for a buck ADC type. Simple design can achieve like 20 000 count accuracy and 60 000 count resolution without a blimpse.

[Rerouter]

Little bit of necro posting, but i believe it was overlooked because you gave no reference to how to implement it other than saying it could be done with jellybean parts

[Yansi]

That is quite obvious, if you know how the dualslope integration works.  Just google it.

You integrate unknown voltage for an exactly known amount of time and then "unintegrate" it down to zero voltage using a voltage reference, but measure the time how long it took.
The measured voltage is then Ux = Uref * T2/T1  where the T1 is the exactly known time (usually selected as an integer multiple of mains period to suppress noise) and T2 being the measured time.

Simple timer and external ISR,  opamp integrator, comparator and a CMOS switch like 4052 is needed. CMOS switch will select what will be put onto the integrators input: Ux, -Uref or COMP_OUT (feedback, that is when you do not need to measure).

There is a ton of possibilities to improve this simple dualslope ADC.

[David Hess]

If the simple ADC requires a whole microcontroller state machine to operate it, then the microcontroller needs to be included in the complexity.  Today though there is a lot to be said for using a cheap microcontroller with its built in 8 or better ADC as a programmable ADC.

For the simplest ADC, I would go with a voltage or current to frequency converter of which there are several configurations.  This is especially the case where a digital output in the form of a frequency is acceptable so no counter is required.  They can be built with a single operational amplifier and no logic to easily achieve better than 8 bits of resolution.

[tggzzz]

The simplest one is the one we had to use at school during physics A-levels, for measurements where the 2% of a moving-coil meter is insufficient.

This ADC give 0.1% accuracy and resolution. It consisted of:

• an ~2V NiFe cell
• a Weston standard cell
• a metre rule with 1mm divisions
• 1m of resistance wire
• a sensitive moving coil meter used to measure the null-point

Put the NiFe cell across the 1m of resistance wire alongside the 1m rule.
Use the Weston cell and null meter to find the point along the resistance wire corresponding to 1.018V. From that calculate the NiFe cell voltage.
Find the null-point along the resistance wire corresponding to the unknown voltage.
Do the obvious arithmetic to determine the unknown voltage.

[Yansi]

If the simple ADC requires a whole microcontroller state machine to operate it, then the microcontroller needs to be included in the complexity.  Today though there is a lot to be said for using a cheap microcontroller with its built in 8 or better ADC as a programmable ADC.

For the simplest ADC, I would go with a voltage or current to frequency converter of which there are several configurations.  This is especially the case where a digital output in the form of a frequency is acceptable so no counter is required.  They can be built with a single operational amplifier and no logic to easily achieve better than 8 bits of resolution.

And  for a V to f converter, you will need what to display the value? Another MCU, huh? Single opamp V to f with "easily 8bit or more", strongly doubt it, no offense intended.
//EDIT: Maybe 8bit resolution, fine, but the linearity will suck!

A simple counter is what you need to operate the integrating ADC.  The state machine in the simpliest version (with continuous intgration) has just two states: Integrating Ux or Integrating -Uref. You do not need an MCU for that, do you? 

What do you think is in the good old 7106? Just a three state FSM of FB/zero, Ux and -Uref and a decadic counter.

[David Hess]

If the simple ADC requires a whole microcontroller state machine to operate it, then the microcontroller needs to be included in the complexity.  Today though there is a lot to be said for using a cheap microcontroller with its built in 8 or better ADC as a programmable ADC.

For the simplest ADC, I would go with a voltage or current to frequency converter of which there are several configurations.  This is especially the case where a digital output in the form of a frequency is acceptable so no counter is required.  They can be built with a single operational amplifier and no logic to easily achieve better than 8 bits of resolution.

And  for a V to f converter, you will need what to display the value? Another MCU, huh? Single opamp V to f with "easily 8bit or more", strongly doubt it, no offense intended.
//EDIT: Maybe 8bit resolution, fine, but the linearity will suck!

The original question did not include anything about a display.

VtoF resolution is usually specified in terms of linearity.  Simple single operational amplifier VtoF converters are easily 10 bits or better.  Drift and PSRR can be challenging.

A simple counter is what you need to operate the integrating ADC.  The state machine in the simpliest version (with continuous intgration) has just two states: Integrating Ux or Integrating -Uref. You do not need an MCU for that, do you? 

Even with provisions for a display, a VtoF converter is simpler.  See below.

What do you think is in the good old 7106? Just a three state FSM of FB/zero, Ux and -Uref and a decadic counter.

And a pile of CMOS analog switches, and a state machine to control them, etc.  If we just want the most integrated converter then the criteria are much different.

I am very familiar with the 7106 and its ancestors.  We always had trouble with dielectric absorption in the integrating capacitor limiting linearity such that the capacitors had to be qualified.  Oddly enough, the 7106 was also sensitive to clock jitter.  I might have liked the LD120/LD121A and LD111A/LD110 which should have suffered less from dielectric absorption but never got to use them.

Remove anything in common and compare what is left over to the schematic of an ancient and admittedly terrible but still 10 bit linear VtoF converter shown below.  Add a counter and timebase for a parallel digital output or display.  Implement the operational amplifier with discrete transistors and it becomes a completely discrete design.

An even simpler VtoF converter using a single operational amplifier to drive an LTC1043 CMOS charge pump can achieve 0.005% linearity for 20,000 counts.  Temperature drift then becomes the chief error.  A more complicated version using two chopper stabilized amplifiers an some clocking logic achieves 0.001% linearity and 100,000 counts by correcting residual charge injection.

[Ash]

Well, the simplest ADC I've ever made was 2 resistors, a capacitor and 2 digital IO pins on a micro..

R1 is the input resistor to Cap. Cap goes to ground. R2 is from output pin to Cap. other pin is an input also connected to the cap.

You then build a simple converter in software.. Been a while, but basically sample regularly (timer interrupt?). If the input is high, drive the output low, if the input is low drive the output high. You count the number of times the pin is high.. Do this enough times and you can output a value. You are basically charge balancing the cap at the logic level threshold of the input pin. Don't expect super accuracy and that drifts and probably has hysteresis.

You can scale the R1/R2 ratio to support wide voltage input ranges.. Some messing around needed to tune the sampling rate etc.

I used this to measure a bunch of pots with an old 8 bit PIC with no ADC.. a long time ago now though.

EDIT: might be that you count the number of times the input is low.. one will get you an "inverse" reading..

Ash.

[David Hess]

Well, the simplest ADC I've ever made was 2 resistors, a capacitor and 2 digital IO pins on a micro..

R1 is the input resistor to Cap. Cap goes to ground. R2 is from output pin to Cap. other pin is an input also connected to the cap.

You then build a simple converter in software.. Been a while, but basically sample regularly (timer interrupt?). If the input is high, drive the output low, if the input is low drive the output high. You count the number of times the pin is high.. Do this enough times and you can output a value. You are basically charge balancing the cap at the logic level threshold of the input pin. Don't expect super accuracy and that drifts and probably has hysteresis.

For the equivalent performance, the discrete version of a ramp converter is more complex than a VtoF converter so they were not commonly used except in the types of application you discussed where very low performance is acceptable.

Below is an example of what is required for a 10 bit self timed ramp converter.  The schematic is mismarked and the ramp capacitor is actually suppose to be 1000 picofarads.  Note the clever method used to achieve 10 bit accuracy with the 2N2222 operating in inverted mode to provide a low saturation reset of the integration capacitor.

An 8 bit implementation still requires the comparator instead of a gate input but the reset circuit can be simplified.  I am not sure if a resistor instead of the current source can be used at 8 bits; I suspect not since a reference is still needed.

I used this to measure a bunch of pots with an old 8 bit PIC with no ADC.. a long time ago now though.

The Apple ][ used this method for its analog controller inputs.  The Atari 2600 did as well for its paddle controllers.  Common potentiometers do not have even 8 bits of accuracy so ramp converters suit them just fine.

[schmitt trigger]

Excellent results given the slow source signal of a load cell.  Wow, they still make it... It's over 27 years old.
https://www.digikey.com/product-detail/en/texas-instruments/VFC110AP/VFC110AP-ND/251222

Burr Brown products were so thoroughly engineered and offered such high performance, that they have stood the test of time very well.
Even nowadays, they are still viable components for high performance circuits.

Speaking of A/D converters, the late Bob Pease which designed National's LM331 VCO, mentioned in one of his many magazine articles that some of the earliest customers were Japanese shipbuilders, which used them to send telemetry from deep inside the ship's bowels to the bridge.

[danadak]

Some more simple variations -


http://www.ti.com/lit/an/snoa627/snoa627.pdf



Regards, Dana.

[Yansi]

I am very familiar with the 7106 and its ancestors.  We always had trouble with dielectric absorption in the integrating capacitor limiting linearity such that the capacitors had to be qualified.  Oddly enough, the 7106 was also sensitive to clock jitter.  I might have liked the LD120/LD121A and LD111A/LD110 which should have suffered less from dielectric absorption but never got to use them.

I have actually built some prototypes of the dualslope ADC few years ago. Used some older MKT axial cap for the integrator and at 20 000 counts I couldn't tell any linearity problems, compared to a HP34401 instrument, that I took as a reference. The conversion results vere bang on what the 34401 showed, just +-1LSB different. I still have the measurement comparison table, for the whole range of 0 to 2.24V (it was 22400 counts top it appears) stepped by a approx. 20mV, that was measured on the breadboard prototype.
Actually, the internal count number was up to full 16 bits, then calibrated and scaled to volts display, with the nominal range of 0 to 2V, 0.1mV resolution.

[mikerj]

Well, the simplest ADC I've ever made was 2 resistors, a capacitor and 2 digital IO pins on a micro..

R1 is the input resistor to Cap. Cap goes to ground. R2 is from output pin to Cap. other pin is an input also connected to the cap.

You then build a simple converter in software.. Been a while, but basically sample regularly (timer interrupt?). If the input is high, drive the output low, if the input is low drive the output high. You count the number of times the pin is high.. Do this enough times and you can output a value. You are basically charge balancing the cap at the logic level threshold of the input pin. Don't expect super accuracy and that drifts and probably has hysteresis.

I've used simple converters like this in the past when the cheap micros didn't have an ADC but there is so much choice now that you'd really need to shave the last cent of a big production run rather than use a part with built in ADC.

[David Hess]

We always had trouble with dielectric absorption in the integrating capacitor limiting linearity such that the capacitors had to be qualified.  Oddly enough, the 7106 was also sensitive to clock jitter.

Used some older MKT axial cap for the integrator and at 20 000 counts I couldn't tell any linearity problems, compared to a HP34401 instrument, that I took as a reference.

We used polystyrene or polypropylene film capacitors but some models from some manufacturers were better than others.  Qualifying them was easy enough but then you have to convince purchasing not to substitute a part not on the qualified list.

The clock jitter problem showed up as flicker noise and just required changing the RC oscillator to an LC oscillator.

The LD120/LD121A and LD111A/LD110 that I mentioned are a run-up integrating converter so they use a much smaller value of integrating capacitance allowing the use of a silver mica part.

[Ash]

Well, the simplest ADC I've ever made was 2 resistors, a capacitor and 2 digital IO pins on a micro..

R1 is the input resistor to Cap. Cap goes to ground. R2 is from output pin to Cap. other pin is an input also connected to the cap.

You then build a simple converter in software.. Been a while, but basically sample regularly (timer interrupt?). If the input is high, drive the output low, if the input is low drive the output high. You count the number of times the pin is high.. Do this enough times and you can output a value. You are basically charge balancing the cap at the logic level threshold of the input pin. Don't expect super accuracy and that drifts and probably has hysteresis.

I've used simple converters like this in the past when the cheap micros didn't have an ADC but there is so much choice now that you'd really need to shave the last cent of a big production run rather than use a part with built in ADC.

Yes, I'd not do it again today without a very, very good reason.

I don't want to be working on any products where the decision is about shaving dollars (let alone cents) off the unit costs.. My products aren't high volume, but have actual profit margins and that is how I like it (got to feed the family, and my Test Gear addiction)   

Ash.

[BrianHG]

The OT asked just for an 8-bit ADC, he didn't ask about speed or linearity.
I wonder if with some ingenuity, something like split power supply, 16 NPN transistors and 8-16 small signal mosfets and a bunch of resistors, maybe a few diodes, no counters, no comparators, no op-amps, no clocks or uC, can we actually create a lousy, but functional analog in, 8 bit digital out?  I bet we can achieve a 4 bit ADC with the above listed components, but can we squeeze an 8 bit code output without going to 256 transistors?

[BrianHG]

As for the counting methods, the easiest way would be a tiny DC motor with a contact reading the revolutions into a uC's counter input.  No comparators, no op-amps, no fancy voltage-frequency converter ICs, an almost all mechanical solution except for the uC counting the motor's RPM.

[David Hess]

The OT asked just for an 8-bit ADC, he didn't ask about speed or linearity.
I wonder if with some ingenuity, something like split power supply, 16 NPN transistors and 8-16 small signal mosfets and a bunch of resistors, maybe a few diodes, no counters, no comparators, no op-amps, no clocks or uC, can we actually create a lousy, but functional analog in, 8 bit digital out?  I bet we can achieve a 4 bit ADC with the above listed components, but can we squeeze an 8 bit code output without going to 256 transistors?

Unless it is specifically mentioned, it is a safe assumption that an 8 bit ADC should be 8 bits linear or have a low enough total error to support 8 bits of accuracy.  Otherwise a lower resolution ADC would have been specified.  Remove the linearity requirement and the simple design with a single comparator or operational amplifier becomes even simpler.

The circuit you are suggesting has been used in the past for things like 1-of-10 DACs which are 3.3 bits and then the ADC was made from 10 simple comparators and some logic to produce a 1-of-10 or BCD output from the thermometer code.

You could make an 8 bit DAC that way if precision resistors are used but then a comparator and the logic for a digital ramp generator or successive approximation register would be needed to make an ADC from the DAC.  This was a pretty common thing to do in the 1980s when integrated DACs plus a comparator and some logic were much less expensive than integrated ADCs.

[free_electron]

pulse width modulator . measure duty cycle with cpu.
you can do it with a 555 if needed ...

[Kleinstein]

A PWM signal and a frequency are still analog signals, so this only a first step towards an ADC.

On can do a relatively simple counter with not so many transistors (AFAIK 2 transistors per Bit). So the VCO and counter version can be reasonably simple. So can be a kind of successive approximation version with an R2R DAC and logic from DTL. It can get away with a low number of parts, but still quite a lot of parts.

The lowest transistor count might be a kind of recirculation ADC similar to that used in some Fluke / Phillips DMMs. Not sure one could do it with the mentioned 16 transistors and few mosfets - by I would not expect much more if you really want too. The result would than be in a serial format, starting with the highest bit and going to lower values. The idea could be charging a capacitor in proportion to the input signal (sample phase) - discharge by a certain amount it if the charge is larger than a threshold and than amplify the charge by a factor of 2 (as good as it is possible) before repeating the process. The tricky part is adjusting the factor of 2, the charge removed and threshold.

[CatalinaWOW]

OP needs to define goals more clearly.  The simplest approach is to use an IC.  It can be as little as one component.  Which can be comparable in size to a TO-3 transistor.  It is also the cheapest.  That is why they sell so well.

All of the old approaches to A/D  V to F, single slope, dual slope can be implemented with discrete components, but the approaches suffer from a combination of size, performance and cost issues.  They are useful as learning tools but little else.

The requirement to not use ICs, but do 8 bit conversion is a little contradictory.  There are few uses for binary representations of analog data outside of ICs.  But there are examples.  One might be if the data to be converted is in an environment not survivable by commercial ICs.  Reactor cores.  Deep bore wells.  Jet engine interiors.  Furnaces.  In these cases an IC is not "simple".  Conversion might be desired to maintain SNR while transmitting the data to a more benign location.  But PWM or V to F conversion without full A/D might satisfy the need.

Knowing the real problem might guide the suggested solutions.

[colinza]

If you are OK with a one bit resolution then you can go for a comparator that compares against VREF/2.

And if you make VREF a triangle wave, you have PWM which is very easy to digitise.

Isn't there a design in use which actually uses a sawtooth wave on a comparator and compares the interrupt generated from that to the timing of the sawtooth wave?

[free_electron]

If you are OK with a one bit resolution then you can go for a comparator that compares against VREF/2.

And if you make VREF a triangle wave, you have PWM which is very easy to digitise.

Isn't there a design in use which actually uses a sawtooth wave on a comparator and compares the interrupt generated from that to the timing of the sawtooth wave?

yes. that's called a slope or dual slope integrator. the a/d converters in multimeters work like that . and they get 6 1/2 or more out of it.
a dual slope converter cancels out all its errors.

what you do is charge a capacitor , for known time with a current derived from the input voltage.
then you discharge the capacitor with a known current and measure how long it took.  the ratio is a direct number for the input voltage.

Multislope converters do not fully disccharge the cap. they pump the input for a known time and pump the reference for a known time. they have different current sources so they can change the ramp speed dpeending on the magnitude of the incoming signal.
and 'track' the input that way. such converters can take thousands of readings a second. Multislope is the basis for the 3478 and 34401 multimeters that have been around since the 80's and are STILL the go-to laboratory machines.

the first chip to work that way was the icl7106/icl7107/icl7109 and the RCA 3161/3162