Glitches on OPA549 output when loaded

Just to make sure it's the output stage, can you change the load and see what the behavior is? Is it different now?

Hi Zidane,

Thank's for your interest into this problem. There has been no reply (besides yours) since my original post so I did much of the investigation by myself. Changing the load changes the output behaviour. As I showed for example, at no load the output is very clean and undistorted at up to 50 kHz (for Vpp <= 5 V). Glitches start to appear at approximately 5 kHz as the load current  (zero-mean sinusoïdal waveform, resistive load ) increases. On my recorded waveforms (and also on that of the simulations), you can see that severe glitches are present when the output current just reaches a little bit more than 1 Amp peak, which is still far from the 10 Amp peak capability of the chip. For capacitive loads, glitches tend to appear in the same way at the top of the waveform after zero crossing of the current. Small capacities cause output instability and need to be compensated adequately, which is a known concern.

Since the past days, I tried to design several compensation circuits to try to alleviate this glitch phenomenon when the Op-Amp is loaded. I came up with some interesting solution that gave good results on TINA-TI, but which I still need to test in practice. If it works, I will post the schematic here in hope that it helps anyone designing with the OPA549. Meanwhile I simulated my circuits with a different Op-Amp, the OPA541, and found that it behaves much better than the OPA549. I ordered some of these chips to see what they really give in practice. I will update this thread as soon as I get concluant results.

Hi Christopher,

Thank-you for this presentation. It was very insightful and I now have a clear understanding of the problem. In this seminar, a definitive solution was proposed to eliminate crossover distortion : the use of a pull-down resistor. I have past the last hours trying to apply this technique to my OPA549 circuit, but it does not seem to work. The seminar presents a simple approach to compute the pull-down resistor value (Rpd), but it does not justify the value of the pull-down current based on the circuit or the Op-Amp at use. I took it to be 200 mA as it was a close value to the 190 mA used in the presentation. The circuit and my calculations for Rpd are as follow :

(15 + 5) / Rpd_max = 200m => Rpd_max = 100 Ohms

(15 - 5) / Rpd_min = 200m => Rpd_min = 50 Ohms

Here are the waveforms I obtained for Rpd values in this range and outside :

It seems that a pull-down resistor cannot solve the problem. I may be wrong however in my calculation of Rpd. It could be that I_pd has to be in the same range as I_Load, as it seems to be the case in the presentation (I_Load = 140 mA, I_pd = 190 mA). Applying this reasoning for the above circuit and a high output current leads to a Rpd of about 5 Ohms and less, which cannot be practical. In this vein, the simulation produces a clean output for Rpd = 3 and R_Load = 2. The seminar was clear and very well made, but the provided solutions do not seem to work for the OPA549.

Hi Christopher,

I tried to join my simulation file in my last post but couldn't find how to do it from the reply form menu. Even now, when I click the "Insert Media" button, browse for my simulation file and click the "Insert" button, the window remains idle and an eventual timeout occurs. Is there another way to join a file that is not a picture?

Anyway, I made some tests with your simulation file and could reproduce the clean output as you mentionned. However, I found it strange that the output was not lagging the input, as observed in my simulation and also in practice. So I raised the frequency to 50 kHz and saw if the lag increased. There was still none. I then set the frequency to 100 kHz and realized to my amazement there was still no lag, as the load remained at 600 mOhms. I then set the input voltage amplitude to 20 V instead of 5 V, and for the same frequency (100 kHz), the output remained perfectly clean with no distortion and again a zero lag on the input (the slew rate 12.6 V/μs, which should have cause great visible distortion). This cannot be the OPA549. If this chip really exists, I would buy many of them for sure!

I then completely removed Rpd to see if there was some glitches as I observed in my simulation and also in practice. There was none for all the tests I performed. The conclusion is simply that the Op-Amp of your circuit cannot be the OPA549. Strange because it is labeled as so. I would gladly share my simulation file with you if I knew how to attach it here, or you can always give me your email so that I can send it to you directly.

Thank's for the simulation by the way, you took all of the parameters I had in my circuit.

Great, I just found how to join a file. We have to click the "Options" tab at the top of the page. Here is my simulation file.

Dear Bastien,

Keep that in mind that neither OPA549 nor OPA541 and LM3886 and LM3875 power OpAmps are suitable for unity gain operation. That is your mistake.

Hi Tanja,

The OPA549 and OPA541 are suitable for unity gain operation. See figure 14 on page 14 of the OPA549 datasheet, and figure 3 on page 7 of the OPA541 datasheet. In both cases a buffered Op-Amp (unity gain) is used as a slave. My practical circuit is almost identical to figure 14 of the OPA549 datasheet. I limit its operation to frequencies lower than 5 kHz, otherwise significant crossover distortion appears. You are right about the LM3886 and LM3875 however, and I knew about this concern. At unity gain, the LM3886's output oscillates at high frequencies. This behaviour is of different nature than the crossover distortion observed for the OPA549 and OPA541.

I've been using the opa549 as an audio amp for the last couple of years with Av=-2 (see schematics ) and my CRO doesn't show artefacts you mention, sinusoidal forms are nice and clean. I didn't try it in unity gain configuration.

Hi Sinisa,

This is interesting. My circuit is similar to yours, where I use a 20k potentiometer instead of your 10k, and a 5.1k resistor instead of your 4.7k. I use two Op-Amps and my pins 4,6 and 8 are also tied to ground. The major difference however is that I don't use a bridge configuration as you do. I use a parallel circuit similar to that of figure 14 of the datasheet. Interestingly, this circuit, when simulated on TINA-TI, doesn't exihbits crossover distortion as only one Op-Amp does when simulated independently. In practice, I also observe less crossover distortion in my parallel circuit (compared to one Op-Amp alone), but not as less as the TINA-TI simulation result shows. In your bridge circuit, the Op-Amps are driven differently and it could be that this helps alleviate crossover distortion. I didn't simulate this configuration but I believe that it coud actually work. As a workaround for a single Op-Amp circuit, I have found that adding a very small inductor (< 50 uH) in series with the output and taking the feedback right after it eliminates much of the distortion. This can also be verified in a simulation. However, I couldn't find a way to implement this effectively in a parallel design. Thank's for circuit schematic by the way.

Hi Bastien,

if you look again at my sch. in previous post you'll notice that it's not a simple bridge configuration (note the nested loops). It's based on SuSy idea (google: Supersymmetry Nelson Pass). You can find more details by looking for: 1994—US Patent 5376899: Amplifier with gain stages coupled for differential error correction.

Hi Sinisa,

Yes you are right, this is not a simple bridge connexion. I had never seen such a circuit configuration before and after some research, I found that the bridge topology matched it best, so I called it likewise. It is very similar though, but you have some additionnal feedback that goes from the two Op-Amp outputs to the signal Op-Amp. I believe it could work very well, but I haven't taken the time to perform an in depth analysis of it. I could also have used a similar topology for my circuit, instead of the parallel configuration, which would have allowed me to lower Vs and operate more at the left of the SOA curve.