Analog isolation circuits: any experience on this?

[mfilippa]

isolation circuits

Hello,

I have an application where I have to interface a high voltage analog signal to a DSP/Computer. I have no problem to step the voltage down to reasonable levels.

Isolation is my main concern. I have tried using HCNR201 and had no success when combining with opamps. I'm looking for a nice response up to 1MHz. I tried the BJT circuit proposed in Agilent App Note and it works fine up to 1MHz, but it is not accurate. On the other hand, I could never get this working with fast opamps.

Does anybody know another coupling/isolation methods for fast analog signals? I read a post regarding VCOs. I know what they are and how they work, but I have never seen a practical circuit working. It would be great if somebody can suggest some ICs to start with.

By the way, the signal is both AC and DC.

Thanks !
Mariano

 

[flatulent]

nve analog isolator

There are a variety of isolation methods. You may be better off using and ADC to digitize the signal and send it across the barrier with an opto isolator type circuit. You then use transformers to get the power to the ADC with a rectifier and voltage regulator.

 

[Sonic]

isolation circuits patient

Flatulent's way is the simplest, I just like to expand on it a little and suggest you use a serial ADC to keep the opto isolators count to a minimum, and a low power isolated DC-DC converter. there are many vendors for this part with a different power and isolation ratings usually 1000V to 4000VAC.

Good Luck

 

[alphi]

analog isolation

Try use signal transformer to isolate and transfer signal.

 

[mfilippa]

vco optocoupler galvanic

I thought about the ADC and digital optocouplers. However, I believe it is a very expensive method for signals up to 1Mhz.

First of all, the ADC should sample at 2Mhz or above, which makes this expensive. Besides, there are not many serial ADC for such speeds. And if I go with parallel ADCs, say 16-bit resolution, I'm going to need 16 fast optocouplers for just one signal. Another issue, if I use serial ADC sampling at 5Mhz, 16-bit resolution, this means I would need a serial output of around 90Mbps. This is way too fast for a simple application.

Now compare this to a basic circuit with 4 transistors and an HCNR201, that has almost the same performance.

alphi: I cannot use a signal transformer with DC signals.

Any ideas on manipulating this signal to couple it through transformers? I examined a board from a Yokogawa Power Analyzer (interfases up to 200A and 1000V) and they are using coupling transformers for all analog signals, including DC signals. I found some optocouplers but they were sending information to the high voltage side. Any idea on how they might be doing this? I can send pictures if you need.

Mariano

 

[flatulent]

analog isolator schem

Some companies make analog type opto couplers. They have feedback on the sending side to predistort the signal to compensate for nonlinearities in the opto diodes.

You will still need to make a transformer isolated power supply to power the sending end of the coupler.

I would not expect great linearity and bandwidth from these items, but you may not need extremely high performance.

 

[alex840]

magnetic isolation circuits

Hello mfilippa,
could you describe more precisely your application:
a) DC to 1MHz or some narrow range?
b) how many bits of resolution your plan to obtain?

In general it is better to digitize(or convert voltage to frequency) primary signal directly at high potential and then to transfer digital signal through isolation (optocoupler or transformer). You will also need DC-DC converter.
Be careful with optocouplers: high isolation voltage is specified only as a test voltage. Optocouplers are subject to aging... Real working voltage is much lower than test one. If you isolation volatge is higher than 1kV, use transformers instead of optocouplers.

-alex840

 

[tanuki]

Use two matched optocouplers. Configure the optocouplers such that one effectively cancels the nonlinearities.

One of two arrangements is common.

1) the signal crosses the isolation barrier (optocoupler) and there is amplified and returned back across the barrier (optocoupler) to be compared to the original signal. (Can be designed to provide a fail safe).

2) One of the optocoplers is used to generate a local feedback for the sending amplifier and the other is used to cross the isolation barrier and is received by a matched gain amplifier.

#2 has a higher frequency response. Example: Vishay IL300
**broken link removed**
(The examples are for 200KHz frequency response).

http://www.vishay.com/optocouplers/opto-linear/

Characteristics of Phototransistors: **broken link removed**

Search the web for "linear optocoupler" and/or "matched linear optocouplers".

 

[mfilippa]

I am already using matched optocouplers. Check the picture I'm sending. It is a simulation of the actual circuit I'm working with. It has a good response to about 500kHz.

This circuit is Ok, BUT :
a) it is not accurate, it has a 3% error in the full scale signal at low frequencies
b) uses optocouplers, with their aging effect and non-linearities.

I tried using opamps, but the bandwidth has been greatly reduced. They are very unstable and I couldn't make them work.

I am interested in a solution as alex840 posted, using transformers for higher voltage isolation. Besides, I believe that it can be far more cheaper than using ADCs.

Alex: how do you digitize a signal to couple it through transformers? I was thinking of FM modulation. I need to trasmit from DC to 1MHz.
I was checking that analog devices have some voltage to frequency converters, but the max. frequency is 3MHz and I think it is not enough to transmit a signal of 1MHz.

Any ideas on this?

Mariano

 

[flatulent]

There is another option. Chop the input signal (multiply it by +-1) so that you get a DSB signal and pass it through a transformer. Synchronously demodulate it on the other side. You can generate the carrier signal on one side and pass it through to the other side with another transformer.

 

[tanuki]

VCO based circuits modulate a local oscillator with the signal ... basically FM. This is transmitted across the isolation barrier (i.e. magnetic or opto coupler). Then a PLL (Phase lock loop) tracks the FM signal and the PLL VCO voltage is the demodulation result. With the appropriate alignments the circuit is stable within the PLL's frequency respone.

Linearity and matching of the VCO's is the primary source of error. This method still relies on a matched response of the two VCO's. If 3% is not good enough ... then I think a FM/VCO isolator will be no better.

I like flatulent's idea.

On the HV (high voltage side) to eliminate the DC offset and thermal drift use a method similiar to a chopper stablized amplifier. Invert the signal and a local voltage reference to create two differential pairs. Time division multiplex the signal pairs to use only one transformer to cross the isolation barrier. Then demultiplex the analog levels to obtain the signal and the reference. The differential levels can be used to correct any offsets and thermal drifts.

Search https://www.sensorsmag.com/
Examples:
**broken link removed**
**broken link removed**
**broken link removed**

Search google using combinations of keywords: "differential signals" "isolation barrier" "isolation circuits" "galvanic isolation"

 

[mfilippa]

Flatulent idea is excellent !

Check the schematic and let me know if you agree with this:

On the HV side:

1) run the signal through two opamps: follower and inverting
2) switch the signals using a fast analog CMOS switch
3) couple it through a transformer
4) Swtiching frequency must be 5 to 10 times maximum bandwidth. If I want from DC to 1MHz, I can use a clock of 10MHz

On the isolated side:

I was originally thinking of using the same scheme to demodulate the signal, but I found that I'm going to need extremely fast opamps to follow the 10MHz switching frequency (check Scan.JPG)

So I finally thought about running the output of the coupled signal trhough a switch and a cap (sort of sample-and-hold) and only take the non-inverted part of the signal (check Scan1.JPG)

Do you agree with this or do you think that demodulation on the isolated side should be done some other way?

By the way, thanks for the sensormag links. I printed them all and am going to read them now.

Mariano

 

[flatulent]

You will have to do a real demodulation. The half wave method will not work.

You do not need the two fast amps on the right side. You can demodulate right after the transformer. You center tap and ground the center tap. Then you switch the top winding and the bottom winding in and out.

Follow that by a low pass filter at 1.2 MHz or so and then your low speed amplifiers.

 

[tanuki]

Consider using standard radio modulation circuits: balanced modulators (analog multipliers) with sin and cos phased carrier and matching coherent demodulating receiver. This might be very cost effective while providing amply bandwidth. You could easily consider a carrier frequency around 50MHz. Standard radio circuits are inexpensive and highly integrated.

Something to think about

Example circuits:

Modulator: **broken link removed**
Demodulator: **broken link removed**

An Error Analysis of the Quadrature Modulator and Demodulator:
**broken link removed**

 

[artem]

What is about fiber optic cable ?

 

[alex840]

transformers:
a) you can digitize signal and pass it through
disatdvantage - you still need ADC
but sometimes it is cheaper to use ADC. It was my reason to ask you about resolution.
b) you can convert voltage to frequency and pass it through
disadvatage - this method primary targeted to work with low speed signals. main concern - linearity and frequency response.
c) you can use modulation (thanks, flatulent!).
disadvatage - at 10MHz you will need to rectify signal. My experience shows that it is not a simpliest task... To reach the same linearity as with, ADC you will need the same complexity of the circuit. (IMHO).

So, I would prefer to use ADC with serial data.

 

[mfilippa]

Excellent ! I never thought of grounding the CT and switch the output. This is an excellent idea. I think this is the final solution. It is very cheap and simple, plus I can synchronize all signals to a same clock, thus saving costs and parts. Thank you Flatulent.

I was a little concerned about saturating the transformer core. If the chopped signal is not perfectly symmetric, there would be an increase in current and flux in the transformer core. It will stop whenever gets to saturation, but then I will have non-linearities. Any thoughts about this ?

alex840: Thanks for the suggestion. ADCs were really tempting for me since I was thinking of manipulating the signals on the isolated side using a DSP. Then, transmitting the digitized signal is more convenient for me. However, I found that almost all serial ADCs lie in the kSps range. For a 1MHz bandwith, I would need at least a 5MSps serial ADC and there are not many available. Another issue is cost. Do you have any experience with fast serial ADCs ? I checked TI.com and they don't have any fast ADC with serial interface.

Tanuki: That is a good choice too (and very cheap too). I will think about it and try to work it out. Unfortunately, I don't have experience on radio frequency or FM modulation, but of course I will give it a try. It is a very good choice.

Thank you all for your suggestions and help. I will post the results, schematics ans PCBs as soon as I build them.

Mariano

 

[alex840]

For high speed serial ADC you can check Analog Devices (www.analog.com).
Another proposal: you can place your DSP and ADC at high potential digitize and compute your signal there and then transfer result of computation through isolation. My experience shows that result of computation in most cases has much lower bandwidth. If it is sow in your case - think about my proposal.

Alex

 

[tanuki]

Transformer saturation: You can either add a DC blocking capacitor or allow a small amount of flux imbalance ... the imbalance will be very small at low power levels.

In your prototype, add the DC blocking capacitor ... then you can short it to study the effect of omitting it.

I think the small non-linearity of the core will cancel any imbalances well before any saturation levels can be reached.

 

[7rots51]

do not invent the wheel again

use ADC + Isolation products from www.nve.com and www.analog.com

they have 4 channel isolators in a chip that is small and low cost.

They used the two different magnetic technology ,you can also use some single channel opto with logic output that works up to 1mbps,but nve works ap to 30mbps

I hate to use hand made trans for this task.