SMPS control loop estability

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
As some of you know, I am trying to build an offline half-bridge SMPS for audio use.
My prototype is almost working, but I have an issue with regulation. I started another thread, but as the problem is now different I have decided to keep them separately for ease of search.

The turns ratio gives a peak secondary voltage around 70V, while I want a regulated output of around 60V.

If I leave the supply unregulated, going to max. duty cycle, the output is too high. If I regulate it "manually" adjusting duty-cycle, (but open loop), the voltage is good for relatively heavy loads, but goes to more than 70V with no load or light load, and load regulation is not too god.
The obvious solution is close the loop, of course.

BUT: as I have it now, it oscillates: at light load, there is a 5Hz, 2Vpp oscillation (not sine, but more like triangular) around the correct set value. As load increases, its freq. increases to, say, 30Hz at 150W.
At some load, it even oscillates badly from almost 0V to 50-60V, I suppose that duty cycle falls to 0 and the controller stops.

I have tried to compensate with no results so far.

I attach the schematics of my current feedback setup. By the moment it is using only the internal SG3525 opamp. I know that the optocoupler diode excitement is very simple and will have some temperature drift, but that's not very important for me by the moment.

Can anyone tell me what's wrong??? Thanks!
 

Attachments

  • control_2.jpg
    control_2.jpg
    58.3 KB · Views: 1,169
I've said many times that this arrangement is not going to work, and I've also explained how to get it right. :dead:

Your output filter requires a zero at 280hz for compensation, but it has to be implemented before the optocoupler, and also two voltage rails can't be sensed in that way.
 
Sorry, Eva. I only wanted to put it to work reasonably well without adding more complication, but if that's not possible...

What's the reason for the requirement that the zero is implemented before the optocoupler? You'll agree with me in that it adds complexity and if that can be done in the control side... much better. But I am sure you have a very good reason.

About sensing both rails, well, it is sensing both of them, but it is true that it has no GND reference, so if the load is not somewhat symmetric it can produce different voltages in each rail.:)

Thanks for your help
 
Rather than being useless as you could think at first, the output ripple voltage waveform is very important for the control circuit, but it must be precisely attenuated and averaged by the pole/zero/pole response of the error amplifier before it goes through the optocoupler, because optocouplers suffer from limited and asymmetric slew rates, delays and storage times, thus distorting high amplitude high frequency signals badly.

Also, placing the error amplifier and the compensation after the optocoupler will allow the optocoupler to saturate and cutoff during transients without the error amplifier being able to track the average value of the output voltage waveform and compensate for it in average.

Concerning output voltage sensing, note that the frequency response of the output filter of each rail changes as load is increased. If you take feedback in such a way and apply a different load to each rail you are likely to get instability because you are summing two different transfer functions that may probably interact producing notches.

Also, even when compensation and output sensing are right, you are going to need a minium dummy load to ensure stability.
 
Please have a look at this reference I have just found. It is a push-pull DC/DC, but the output voltages are very similar and in terms of control loop it is almost identical.

http://valveaudio.tripod.com/images/membuat/Gambar9b1012a.jpg

They use a fb. scheme very similar to mine, with no additional opamps. Eva, what do you think of this arrangement? Perhaps it can't work properly?

The only compensation is a pole at the output of the error amplifier. I really don't know how it works. It is wired as a comparator instead of an amplifier with moderate gain.
 
That link doesn't work, this site doesn't allow to access their images from anywhere except their own links.

EDIT: I've managed to reach the schematic, but, does it work? is it stable? :D Who knows? Note that they are using the optocoupler in a very different way.
 
Well, in fact they are using the optocoupler in a very similar way to what I do. (although they put the output voltage directly to the error amp used as a comparator).

I have been reading the SG3524 datasheet (very similar to SG3525) and they recommend compensation by means of a capacitor+resistor from COMP to GND.

I still don't understand why compensation should be done before the optocoupler. Please could you explain me that, Eva.

Sorry for so many questions but I think that this issue is interesting for a lot of people too. Thanks for your time!
 
Feedback Compensation

Pierre/Eva,

Some years ago, I built a DC-DC PWM converter using the SG3525 chip, with compensation on the primary (control) side of the ckt. This was from the Audio Amateur 04/88 and 01/89 issues from an article by Mr. Randy Vikan and his 75+75W all-MOSFET Car amplifier. Pt I was about the SMPS, and Pt II- the Amp itself.

In this schematic, he uses a parallel r-c network (49.9kW and 1000pF), going from the comp pin (pin 9) of the '3525 to ground. Switching frequency was just over 26kHz, so I upped this to 33kHz, further removing it from the audible spectrum. This didn't seem to make any difference in the performance of the converter from a noise perspective. Though I didn't have a ScopeMeter at the time, audible inspection and voltage measurements verified this. The optocoupler, an MCT271, is no longer made and is replaced by the garden-variety 4N36 or similar optocoupler. "Compensating" for the variations of CTRs (Current Transfer Ratios) in different optocouplers is a biasing network, using the very-familiar TL431 and a few resistors. I believe the '431's reference pin taps a 61.9kW (upper) and 6.19KW (lower) resistor divider chain that sits across the +/-37V Buses. From the +37V line comes a 2.21kW resistor, through the LED portion of the optocoupler, and to the Anode of the '431. The '431's cathode simple dumps to the -37V line. Sensing was about the +/- buses, with no reference to the output side's groung node. BTW, input and output grounds are isolated from each other, coupled only by a couple of 0.01uF caps.

Anyway, I have built this DC-DC Converter with little-to-no noise on several discrete and chip- amps. This is used as simple voltage-mode, no current-mode sensing is used. That is not to say there may be some advantage(s) in implementing current-mode sensing with the '3525 (yes, it can be done with this chip), but, since this was a pre-made board, I went with what was made space for.

The only difference between my version and the author's was the substitution of two FT-140-77 toroid cores (Amidon), stacked, in place of the single 3622 pot-core (Ferroxcube). If I can find a web-based schematic, I will post it here.

Going a step further, I transformed the '3525's control circuit into a "modularized" half-bridge off-line SMPS, and powered a pair of 40W discrete Amps, with excellent results. I kept the same feedback compensation scheme, and achieved good results. Unfortunately, I have since disassembled the circuit for use of the parts in other circuits. Boy, am I wishing that I didn't do that! :bawling:

Anyway, this is getting long-winded, so I will close by saying, that, in my experience, compensation on the control (primary) side with the 3525 is indeed possible. Pierre, can you go back and check your parts layout? You'll be suprised how much this can affect stability.

All the best,

Steve
 
Hi, N-Channel.
Nice to hear that.
Althouth Eva's proposed approach will undoubtfully work better, I still think it is possible to operate SG3525 without additional opamps in a estable way.

Let's talk a bit more about the circuit you talk about:

+VCC (positive rail output) goes to a resistor, then photodiode anode, then TL431 _cathode_, then -VCC (negative rail output).
The terminal control of TL431 has a 69K resistor to +VCC and a 6.9K resistor to -VSS.
Is this correct?

Thus, if voltage increases, ref. voltage will increase and so TL431 zener voltage, so current through the LED will decrease.

So... the control side should be something like that:
Phototransistor collector to +12V, emitter to GND with around 1K resistor in series. The error signal is taken from the emitter, and goes to the SG3525 error opamp, configured as a simple unity-gain noninverting buffer, right? (inverting input to COMP, + input to phototransistor emitter.

COMP pin with R||C to GND (49.9k and 1000pF)

Is all this correct or is it wired in a different way? An schematics would be great.
 
Pierre,

A few corrections on your last comment:

So... the control side should be something like that:
Phototransistor collector to +12V, emitter to GND with around 1K resistor in series. The error signal is taken from the emitter, and goes to the '3525 error opamp, configured as a simple unity-gain noninverting buffer, right? (inverting input to COMP, + input to phototransistor emitter.

COMP pin with R||C to GND (49.9k and 1000pF)

Is all this correct or is it wired in a different way? An schematics would be great.


The (+) error amp input of the '3525 is connected th the junction between phototransistor's emitter and a 4.99kW resistor. The resistor goes to GND, and the phototransistor's collector goes to the Vref pin (pin 16) of the '3525, which puts out 5V up to 50mA.

The (-) error amp input of the SG3525 is connected to the junction of a pair of 4.99kW resistors. One resistor goes to GND, and the other goes to Vref.

The output pin of the '3525's error amp goes to ground, via the 49.9kW resistor and 1000pF cap in parallel.

I hope this clears up your confusion. Your understanding of things on the secondary side of the opto are correct.

Putting two 4.99kW across the +5V reference voltage gives a bias current of 500mA, for the inputs of the error amp.

There is a complete design procedure for sensing the output voltage of isolated DC-DC Converters the book titled "Power Supply Cookbook", by Marty Brown, ISBN: 075067329X, EDN Series for Design Engineers, Motorola Series in Solid State Electronics, Butterworth-Heinmann, MA, 1994, 248 p. Look on p.75 of the Revised 2001 edition of the book

His book uses a cut-n-dried approach to properly biasing and compensating feedback circuits on either the primary OR secondary side, depending on the topology used (flyback -v- forward), and whether the converter is output-isolated or not.


Steve
 
A lot of thanks for the clarifications, N-Channel.
I am going to test that. Let's see how it works.

A couple of things more: in your previous post, I think you have put the TL431 polarity backwards. The cathode should go to the cathode of the LED while the anode goes to -37V.

Did you test this supply with light loads or no load at all? Did the output go higher than being loaded? I am particullarly worried about that because I plan to set the voltage to +/-60V, but don't want it to go higher than that with no load as I am using 63V capacitors and the amplifier can be damaged also when no audio power is produced and hence current drawn from the supply is very low.
 
After thinking a bit more about N-Channel proposal, I see one problem when output voltages are high (say, +/-60V), as TL431 only accepts 36V, unless the resistor in series with the photodiode is absorbing the excess voltage drop.

However, there is still one thing I don't understand about the secondary part of the circuit:
TL431 works as a "variable zener" whose voltage is V=Vref*1+(R1/R2), when R1 is connected to the cathode to the adjust, and R2 from adjust to the Anode. (according to the datasheet basic application).
But, in the proposed circuit, it is not wired that way, so, how can I calculate V between cathode and anode as a function of output voltage and the resistors?

If I knew that, I could dimension the series resistor accordingly so LED current is ok and voltage rating of the TL431 is not exceeded.

Sorry if that's a very basic question but I am a bit confused.
 
Pierre,

You're right about the TL431. I mis-described its connections backward. However you describe backwards, the anode-cathode connections of the TL431's LED.

I'm not too sure about the resistor in series with the opto's LED, as I do not have the design procedure in front of me, but I think it goes something like this:

For a 500uA biasing current at the error amp inputs with an optocoupler having a 100% CTR, the biasing current for the opto's LED ahould also be 500uA, so size the resistor for a current of 500uA.

The TL431's Vref is 1+(R2/R1), for R2 being the upper bias resistor, ands R2 being the lower bias resistor. So, for 74V (+37V-(-37V)), the Vref of the TL431 is 1/11th of this voltage difference, or (74V/11) = 6.72V above the -37V bus, or -30.27V.

So, for a current of 500m[/FONT=symbol]A sinking into the '431 from the +37v line, we get 37-(-30.27) = 67.27V. Now take 67.27V - V(LED - opto) = 67.27-2.2V = 65.072V.

And since R = V/I, 65.072V/.0005A = 130K. For 1.0mA of bias current, make the series resistor 65.072K. (Use 66k as the next std value). Obviously, for your application of +/-60V, your values will be different- this is just an example.

Anyway, these are for optos with 100%CTRs. If yours is lower, say 50%, then a 2.0mA LED bias will yield 1.0mA at the error amp input. Also, the two 4.99k resistors tied to the other error amp input will now be 2.50k across the 5Vref of the SG3525. Taking a CTR of 50%, the biasing resistor should be 32.5k. Use 33.2k as the next standard 1% value.

This arrangement keeps things below the TL431's 36V limit.

Hope this Helps.

Steve
 
No Problem! :cool:

After going back and reading my last post, I discovered some minor math errors, but the general theory holds. Also, I should have said thar R2 is the upper resistor, and R1 is the lower resistor. I acciidentally listed R2 as both upper and lower.

Here is a pic of the DC-DC converter I built. More to follow......

Regards,

STEVE :cool:
 

Attachments

  • rkv dc-dc other side.jpg
    rkv dc-dc other side.jpg
    74.4 KB · Views: 612
Sg3525

...last one!

These are old parts. Note the Motorola symbol on the '3525. In fact, all the semi's are Motorolas, the PWM IC (SG3525AN), the N-Channel MOSFETs (IRF540), the output rectifiers (MBR10010), the Optoisolator (MCT-271), and the 19V 1W zener diode (1N5???), the TL431, and the reverse-polarity protection diode (1N5406).

On subsequent versions, I will substitute the MCT-271 with the CNY17-2, which has a CTR ranging from 63-125%, or the 4N35, which is steady at 100%.

Also, note the pot I put on the very edge of the board- this allows me to tweak the output voltages +/- a few volts.

Also, the biasing resistor has been lowered from 36.5k to 33.2k, yielding voltages for the amp's output section at +/-33V, and voltages for the amp's input stages at +/-42V. This tiered output arrangement is necessary for the amplifier's output MOSFETs, which require a voltage higher than the Vdd and lower than the
-Vss to operate over their entire linear region.
 

Attachments

  • rkv sg3525.jpg
    rkv sg3525.jpg
    70.6 KB · Views: 534
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.