• Resolved

LM675: High Frequency Output Oscillations

Part Number: LM675

Basic Description:

I am having a tough problem with high frequency output oscillations with an LM675 (and also an LM1875) on 2 of the same PCBs, call them PCB1 and PCB2.  The relevant schematic excerpt is attached.  The normal loads connected via J20 are a 10ohm resistor in series with a 10uH-1mH linear (non-saturating) inductor.  Nominal drive current range is +/- 500mA, and I need DC-10KHz output.

PCB Layout:


The current PCB layout is attached.  I've also attached a previous PCB layout in which the problem did not occur!  The differences (current vs previous) include:

  1. Current - 4 layers 2oz with GND, V+, V- planes, vs Previous - 2 layers 0.5oz with only GND plane
  2. Amplifier mounting and bypassing
    1. on the current, then amplifier is mounted standing up vertically into through-holes.  The bypass caps are directly underneath it on the bottom side (only components on bottom side).  I accidentally forgot to void the power plane connections inside the THs for the amp's power pins, so they are connected to the power planes both directly and through a via/trace on which the bypass cap sits.
    2. on the previous, the amplifier is laying down horizontally with pins trimmed and bent down onto SMT pads.  Bypass caps sit on the same top side, with power traces hitting the bypass caps right before they reach the amp pin pads themselves

Description of problem:

The problem I am seeing is a very high frequency (5-30MHz) large (>1Vpp often) oscillation on the amplifier output.  See attached.  This is at and above the amplifier GBW.  This oscillation appears only at certain input voltage (and thus output current) levels, which is really really odd to me.  The oscillation frequency is different depending on the I/O polarity, say typically ~25-30MHz on + side and closer to ~5MHz on the - side.  It will go away if the output current magnitude is too high or too low, it has a sort of sweet spot around 50-150mA in both directions.  I've messed around with the input attenuator to change the Vin / Iout relationship and it seems to follow the output current values rather than the input voltage values.

Testing performed:

I have performed the following testing on PCB #2 (the one with the LM1875 rather than the LM675) since it is not installed in it's fixture and the symptoms between the 2 were quite similar.

  1. Apply small capacitor (33-68pF) across J20.1-J20.2 - often greatly improves, but doesn't completely eliminate.  Not totally consistent.
  2. Remove 6.8u tantalum bypass caps - oscillations get worse (grow in amplitude)
  3. install 33u ceramic caps in their place - negligible effect
  4. replace inductive load with short circuit (leaving resistive load in place) - negligible effect
  5. shorten load connection leads from ~16" to ~3" - negligible effect
  6. Apply 1x scope probe 3" from amplifier output - quells oscillations almost completely (acting as output to GND snubber?)
  7. Apply 1x scope probe directly near amplifier output - negligible effect or makes worse
  8. Apply discrete snubber, 22R series with 100p:
    1. if series inductance is within 150nH-400nH, quells oscillations (small values furnished by straight wire, wire loops, etc, very consistent results.  places resonance near ~20MHz ballpark)
    2. if series inductance is outside this range, negligible effect
  9. Replace LM1875 amplifier with LM675 - quells oscillations completely
  10. Replace that LM675 with another LM675 - quells oscillations completely
  11. Move LM675 from top mounted, bottom mounted, top mounted again - oscillations always gone

Note that PCB1 exhibits the same issue as PCB2 and it is already using an LM675... 

Some questions:

  • Does anyone see anything criminal about my new PCB layout compared to the old one?  I would really be willing to re-spin the PCB's if I had half a clue of what change I'd even make...
  • Is there any reason to expect that an LM675 would behave differently than an LM1875 in this regard?  I am still scratching my head at the concept that that fixed it on PCB #2, despite not obviously helping on PCB #1
  • Am I just stupid for not including an amp output to ground snubber ckt?  I see them on a lot of designs but I didn't include one, and it doesn't really seem like my problem is a normal load related feedback instability.  The finnickyness of the snubbers in my case (150-400nH series inductance requirement) does not make me love the idea.  I'd rather remove the oscillation root cause...

Attachments:


  • I am going to address your issues in 2 separate replies. First a couple of key tutorials on op amp stability:

    Op Amp Stability Link:
    e2e.ti.com/.../2645.solving-op-amp-stability-issues

    V to I Circuits Link:
    e2e.ti.com/.../3600.high-current-v-i-circuits
    Tim Green, MGTS
    Senior Analog Applications Engineer
    
    		
  • In reply to Tim Green1:

    There are many variables here so we should eliminate them 1 by 1.  

    1) The snubbers are a must located directly at the op amp output to a single point ground.  I suspect there is some cabling to our load off of J20, which will be capacitive.  Use the datasheet recommended 1ohm and 0.22uF for now.  At the end of this post I attach a presentation on the "All NPN Output Stage" which is used on both of these power op amps.  

    2) You must keep the gain across frequency on these op amps >10.  That means NO Feedback capacitors.  If we need to there are other tricks to stabilize the circuit.

    3) Once you change to current mode you will need to consider compensation for stability covered in the presentation in my first post "High Current V-I Circuits".

    4) The output stages of these devices are wider bandwidth that the amplifier itself.  The All NPN Output can oscillate due to local oscillations at a frequency > UGBW.  Since your oscillations are >UGBW, if you keep pure resistive feedback at G>10, it is possibly an output stage oscillation that the snubber in 1) should fix.

    5) Another variable to eliminate is thermal issues.  I do not see any mention of heatsinking?  If heatsinking remember the metal tab is tied to -VEE so tab must use insulating washer or heatsink must float or be tied to -VEE.  We can easily do a power dissipation calculation simulation based on voltage input at IN+, load on output (resistive, inductive, etc), assume x10 gain, no heatsinking (?), ambient temperature.  This would predict junction temperature.  High junction temperature can cause parameters to be different.   Reactive loads can cause the op amp output stage to dissipate power that is not obvious at first glance.

    6) Cannot tell you why LM675 looks different than LM1875 since LM675 does not have a lot of analog plots I its datasheet.  These designs were done circa 1999 (LM675) and 2004 (LM1875) so a lot of detailed internal design information is not easy to find.

    7) The oscillation occurring under different loads and output operating points is not easily explained but not uncommon if we have the output stage oscillating as changing operating currents in the output stage and load phase shifts will add to the issue.

    OPA2227 All NPN Output Analysis.pptx

    Tim Green, MGTS
    Senior Analog Applications Engineer
    
    		
  • In reply to Tim Green1:

    Hi Tim,

    Thank you very very very much for the detailed reply.  I have learned a ton just from this post, but one thing sticks out hugely: G>10 across frequency, no feedback caps.  Holy cow. I ended up doing some further testing and found out that removing C43 (the pure 100pF feedback cap) almost completely solves the problem on another test board.  It still rears up right around 50mA in certain conditions but it is significantly improved, and that is still with the 100pF / 560R series pair in feedback which perhaps I shouldn't have either.

    (I did test this previously and for some reason did not see it making a difference, but I must have had some other conflating changes at the same time.)

    For completeness see my comments to your points below:

    Tim Green1

    There are many variables here so we should eliminate them 1 by 1.  

    1) The snubbers are a must located directly at the op amp output to a single point ground.  I suspect there is some cabling to our load off of J20, which will be capacitive.  Use the datasheet recommended 1ohm and 0.22uF for now.  At the end of this post I attach a presentation on the "All NPN Output Stage" which is used on both of these power op amps.  

    I will try this out 

    2) You must keep the gain across frequency on these op amps >10.  That means NO Feedback capacitors.  If we need to there are other tricks to stabilize the circuit.

    I can see now that this is absolutely key, as above.

    3) Once you change to current mode you will need to consider compensation for stability covered in the presentation in my first post "High Current V-I Circuits".

    4) The output stages of these devices are wider bandwidth that the amplifier itself.  The All NPN Output can oscillate due to local oscillations at a frequency > UGBW.  Since your oscillations are >UGBW, if you keep pure resistive feedback at G>10, it is possibly an output stage oscillation that the snubber in 1) should fix.

    5) Another variable to eliminate is thermal issues.  I do not see any mention of heatsinking?  If heatsinking remember the metal tab is tied to -VEE so tab must use insulating washer or heatsink must float or be tied to -VEE.  We can easily do a power dissipation calculation simulation based on voltage input at IN+, load on output (resistive, inductive, etc), assume x10 gain, no heatsinking (?), ambient temperature.  This would predict junction temperature.  High junction temperature can cause parameters to be different.   Reactive loads can cause the op amp output stage to dissipate power that is not obvious at first glance.

    I am using a pretty large heat sink on the amplifiers, and I have in fact noticed an effect of junction temp on the behavior.  On this latest test board, in which the oscillations are almost gone but still appear slightly right at a certain output current (~+50mA), a freshly powered up amplifier will exhibit worst oscillations and improve (eventually die) as it heats up. Without a heat sink, takes ~5s, with a heatsink, much longer.  Cooling the unit with air slowly brings them back.

    6) Cannot tell you why LM675 looks different than LM1875 since LM675 does not have a lot of analog plots I its datasheet.  These designs were done circa 1999 (LM675) and 2004 (LM1875) so a lot of detailed internal design information is not easy to find.

    That's fair enough.  Despite much being the same, their Q currents are actually quite different (~18mA nominal vs ~50mA nominal?) so maybe whichever part of the circuit heavily consumes bias current (output stage?) is operating quite differently between the two.

    7) The oscillation occurring under different loads and output operating points is not easily explained but not uncommon if we have the output stage oscillating as changing operating currents in the output stage and load phase shifts will add to the issue.

    (Please visit the site to view this file)

  • In reply to Dominic Ryan87:

    I sometimes hear a customer say that his circuit has an acceptable level of oscillation. This makes me very concerned as any level of oscillation should not be acceptable. Oscillation at any output current is a warning sign something is not right. I did not go through your layout in detail but the rule of thumb to keep in mind is that op amp inputs are high impedance so keep low impedance traces (like op amp outputs) and other low impedance sources away from op amp inputs.
    Tim Green, MGTS
    Senior Analog Applications Engineer
    
    		
  • In reply to Tim Green1:

    Thanks for the information.  There a few spots in the layout where perhaps I could observe this rule of thumb better, so I will have to check them out.

    If you don't mind, I have a few questions about the snubber circuit, since these are really odd to me having come from a low-power / signal type design background.  Even brief answers to these would be quite helpful.

    1) I see that 1Ohm / 0.22uF is the preferred value set in the datasheet.  Can you explain how these values would be chosen?  I noticed that the R and C impedances cross around ~720kHz which is ~10x higher than the power bandwidth.  Is that the figure of merit?

    2) If I want to maintain the ability to have large voltage swings at fairly high frequency I would need really beefy components at 1 ohm / 0.22uF.  Is there likely any harm in scaling to 10 Ohm / 0.022uF?

    3) Is it safe to assume low ESR caps (e.g. ceramic) are the best choice for this? I would assume so since the explicit R is quite low at 1 Ohm

    4) Is it fair to understand the operation of this circuit as effectively loading down the output so much at higher frequencies that the output stage cannot produce enough current quickly enough to create the voltages that would begin and sustain HF oscillation?  Something like putting a big fat weight on a high Q mechanical resonator to stop it?

    Thanks again for your help.

  • In reply to Dominic Ryan87:

    Hi Dominic,

    Tim is out of office for the next few weeks so I will try to provide some additional insights.

    1) It's difficult to determine how values are chosen for this device because as Tim mentioned we do not have much data on it. I believe the values in the datasheet are intended to be outside the closed loop bandwidth you would need but within the bandwidth of the NPN output stage loop so as to attenuate higher frequency signals that might produce oscillation in the output stage.

    2) Increasing the output resistor may not yield as good of performance because it results in less loading at high frequencies and pushes the pole and zero in the Aol curve created by this snubber closer together and closer to the pole introduced by your load. This compensation method works by effectively having the RC on the output appear like the dominant load at higher frequencies, but if your actual load starts to approach the same value then the method will not be as effective. I think this is going to require some trial and error to verify. I can't intuitively say whether 10 Ohms will be acceptable or not.

    3) Your assumption is a good one. Given the low resistor values (used because of the very low output impedance of power op amps), low ESR caps would be ideal, otherwise your compensation is at the mercy of the ESR.

    4) Yes I think this is a good way to think about it. You're essentially weighing down the output with a low impedance path to ground at the frequencies of concern to kill the gain of the output stage.

    Regards,

    Zak Kaye
    Precision Amplifiers Applications 

  • In reply to Zak Kaye:

    Dominic

    We haven't heard back from you so we assume this resolved your issue. If not, post another reply below.

    Thanks
    Dennis
  • In reply to Dennis Goeke:

    Hi Dennis,

    I am confident it will but I haven't gotten to test yet, could I hold off until I can post some success shots? Just in case
  • In reply to Dominic Ryan87:

    Hi Dominic,

    you asked:

    "1) I see that 1Ohm / 0.22uF is the preferred value set in the datasheet. Can you explain how these values would be chosen? I noticed that the R and C impedances cross around ~720kHz which is ~10x higher than the power bandwidth. Is that the figure of merit?

    2) If I want to maintain the ability to have large voltage swings at fairly high frequency I would need really beefy components at 1 ohm / 0.22uF. Is there likely any harm in scaling to 10 Ohm / 0.022uF?"

    You can understand the effect of snubber (Baucherot-, Zobel-network) only if you keep in mind that the output stage of OPAmp (LM675) has a finite open loop output impedance, Ro. At unity gain frequency Ro and R of snubber must form a resistive voltage divider, to do two things: Decreasing the gain of OPAmp so that a lower portion of output signal is fed back to the inverting input, and doing this without introducing an additional phase lag.

    You mentioned that the corner frequency of snubber is at arround 720kHz. This is to make the snubber look "pure" resistive at the unity gain frequency of 5.5MHz of LM675. So, the corner frequency of snubber must not be changed!

    And also the value of R of snubber must not be changed, because the voltage dividing in combination with Ro will not be sufficient then.

    Be very careful, some OPAmps need this snubber to work stable. It's part of their frequency compensation. This is the case for the LM675. You won't see any figure or example circuit in the datasheet where this snubber has been omitted...

    Kai
  • In reply to kai klaas69:

    kai klaas69


    You can understand the effect of snubber (Baucherot-, Zobel-network) only if you keep in mind that the output stage of OPAmp (LM675) has a finite open loop output impedance, Ro. At unity gain frequency Ro and R of snubber must form a resistive voltage divider, to do two things: Decreasing the gain of OPAmp so that a lower portion of output signal is fed back to the inverting input, and doing this without introducing an additional phase lag.

    This does make sense, thank you.  


    You mentioned that the corner frequency of snubber is at arround 720kHz. This is to make the snubber look "pure" resistive at the unity gain frequency of 5.5MHz of LM675. So, the corner frequency of snubber must not be changed!

    I confirmed this looking at the bode plot of the network, the phase really approaches 0deg around then.


    And also the value of R of snubber must not be changed, because the voltage dividing in combination with Ro will not be sufficient then.

    Unfortunately I have no choice but to disobey this in my application, because I need to reserve large signal swing at high frequency which would cause absurd power dissipation in the snubber with R=1 / C=0.22uF.  However at least in my specific situation scaling to 10R / 22nF (as to preserve the corner) seems to work for what I'm doing.  See post below:


    Be very careful, some OPAmps need this snubber to work stable. It's part of their frequency compensation. This is the case for the LM675. You won't see any figure or example circuit in the datasheet where this snubber has been omitted...

    This is true and I feel like a fool for having ignored it to begin with now

    Kai