@TeeKay6,

I'm not going to have time to do item 6 (extreme cure) today. I'll do that tomorrow. I've done 3 (trace of C2) and 5 (remove board from grounded lid). Traces are in the attchment LogBook-Vol06-NoninvOscillationV3.pdf fig40 and fig41. The ground plane removal made no difference. There's some photos fig42, and a bit more info in the attachment.
Best regards
Dave

@Dave Lowther
Thanks for the update. A couple more things that we now know are not significantly involved in producing the oscillation. Thus far, all data indicates simple capacitive coupling from pin1 to pin3 leads & components.

Another thing to consider: If in normal use you will be using coax at JP1, it would still be worth knowing how much capacitance from JP1:1 to :2 (without coax) is required to kill the oscillation. How does that value compare to the total capacitance of the coax you intend to use to connect to JP1?

Assuming that we ultimately eliminate the oscillation, I did notice a couple of graphs in the LM4562 data sheet (Fig107, 108) that indicate some sensitivity to even light capacitive loading on the output (pin1); I believe that adding a resistor in the connection from pin1 to JP4 is warranted; perhaps 100Ω (short lead to pin1), value dependent on how connection is made to next stage from JP4 and loading by next stage. This also indicates that care must be used in determining connection of the next stage.

 

@TeeKay6

@Dave Lowther
6. If feasible, lets try an extreme cure case. Connect 0.1uF bypass caps directly (i.e. through the air but leads as short as possible) to the ground end of C2 (no need to remove existing bypass caps). Disconnect everything from pin3, at pin3, and then add a 470K-1M resistor (very short leads) from pin3 to the grounded end of C2.

When you say C2 above, I think you mean C5. I don't see how I could have very short leads on a resistor going from pin3 to the grounded end of C2 (see fig37). I'll proceed on the assumption that you meant C5.

 

@TeeKay6
Having done the changes for step 6, it didn't stop the oscillation but it did shift it up to 37 kHz.
More details are in the attachment LogBook-Vol06-NoninvOscillationV4.pdf (from page 35) including a diagram showing where tracks needed to be cut, photos, and a trace of the output oscillating.

Now I'm thinking of experimenting to see:
a) What gain is high enough to cause the oscillation (substitue higher values for R4, or use a preset) with the input open circuit
b) What value of resistance across JP1 is high enough to cause the oscillation at various gains.

You asked what type of cable would be connected to the input. Typically it would be a scope probe. I note what you said about having e.g. 100R in series with the output pin. I was aware of this, I mentioned it way back in comment #3.

I'm now thinking there must be a better (more stable) way to make a high impedance input amplifier with a gain of 100. I'm wondering whether I'd be better having a high impedance voltage follower (gain 1) followed by a low impedance amplifier with a gain of 100. Perhaps the gain split doesn't need to be 1 + 100, maybe it could be e.g. 10 + 10. Perhaps it would still be a problem with coupling from the output of the 2nd stage back to the input of the 1st stage. Perhaps I should use separate chips for each stage.

I did previously experiment with 2 stages. The reason for switching to 1 stage was to see if it would reduce THD and noise.

I'm going to take a step back and have a think about whether it's worth pursuing the one stage solution.

Thanks again for all your help.
Dave

 

@TeeKay6

When you say C2 above, I think you mean C5. I don't see how I could have very short leads on a resistor going from pin3 to the grounded end of C2 (see fig37). I'll proceed on the assumption that you meant C5.

@Dave Lowther
I did mean C2. For the "extreme" test I am advocating a version of the single ground concept. Most especially, this means that all bypasses and feedback components connect to one point. You can actually pick whatever point you wish, although the closer to IC1A the better. Whatever connects to ground does so at one point (or as close as feasible; it doesn't matter for this test whether you could design a nice looking breadboard with this test ground condition; wires thru the air are fine, no need to follow board surface). For the test we are removing all non-essential components. The goal is to see if the amp is stable (not oscillating) in that ideal condition.

 

@TeeKay6

@Dave Lowther
I did mean C2. For the "extreme" test I am advocating a version of the single ground concept.

I think I'm close to having a single ground at the ground end of C5. I'll have another look at it tomorrow and see what else I need to re-wire to use the ground end of C5 as the single ground point.

[EDIT] I've had a quick look and I think I just need to remove the connection between R4 and C2 at R4, then connect another C2 from the same end of R4 to the ground end of C5.

[EDIT2] On 2nd thoughts I think I should remove the connection between R5 and R4 at R5, then add a new C2 and R4 with short wires using the ground end of C5 as the ground.

Best regards
Dave

 

@Dave Lowther
I did mean C2. For the "extreme" test I am advocating a version of the single ground concept. Most especially, this means that all bypasses and feedback components connect to one point. You can actually pick whatever point you wish, although the closer to IC1A the better. Whatever connects to ground does so at one point (or as close as feasible; it doesn't matter for this test whether you could design a nice looking breadboard with this test ground condition; wires thru the air are fine, no need to follow board surface). For the test we are removing all non-essential components. The goal is to see if the amp is stable (not oscillating) in that ideal condition.

@Dave Lowther
You have a compulsion to neatness that is not necessary for tests, since they can be disassembled as soon as the test is complete. However, in general, neatness alone will not invalidate a test. And you do have a very neat "paper trail" of your work. However, neatness does not in general lead to better circuit performance.

I have not traced out your test wiring; I will accept that you have met the requirement of a single ground point with all unnecessary components disconnected from the amp. Your stated result is that the amp still oscillates and I assume that means the waveform is still a rough squarewave--i.e. oscillation is not marginal but quite dominant. As I previously stated, either we have missed some significant routing or connection OR we will never succeed in stabilizing the LM4562 on your board. (Many high speed circuits can be realized only via actual PCBs, not breadboards.) Since the most basic configuration is unstable, I have nothing more to offer on that topic.

Yes, it is certainly possible that you will have more luck using 2 stages. Clearly the stumbling point is achieving an amplifier (configured for some closed-loop gain that is TBD) that tolerates an open input. Because of our experience with the LM4562, I am inclined to keep each stage in a different chip so that physical separation can be achieved; that is, we can keep the scope probe input as well separated as possible from the output of the 2nd stage. Separation is difficult to achieve in a single IC package. It may be that simply preceding the LM4562 with a simple unity gain buffer would be sufficient as that would allow driving the LM4562 from a low impedance.

I had considered building a DIY differential scope probe some years ago, but after a bit of study I decided that I had little chance of achieving wide differential bandwidth (at least 10MHz) and high common mode rejection (at high frequency); that's one reason why commercial differential probes are so expensive--they are not easy to design. Also, beyond a few exploratory tests, each design iteration would require fabricating a new PCB; breadboarding would not suffice. A dozen or two PCB iterations could get expensive and I have neither skills nor experience in RF design.

You could also consider whether requiring the LM4562 to be stable with an open input is reasonable. In actual use, surely you would always have a probe connected.

 

@Dave Lowther
You have a compulsion to neatness that is not necessary for tests, since they can be disassembled as soon as the test is complete. However, in general, neatness alone will not invalidate a test. And you do have a very neat "paper trail" of your work. However, neatness does not in general lead to better circuit performance.

I have not traced out your test wiring; I will accept that you have met the requirement of a single ground point with all unnecessary components disconnected from the amp. Your stated result is that the amp still oscillates and I assume that means the waveform is still a rough squarewave--i.e. oscillation is not marginal but quite dominant. As I previously stated, either we have missed some significant routing or connection OR we will never succeed in stabilizing the LM4562 on your board. (Many high speed circuits can be realized only via actual PCBs, not breadboards.) Since the most basic configuration is unstable, I have nothing more to offer on that topic.

Yes, it is certainly possible that you will have more luck using 2 stages. Clearly the stumbling point is achieving an amplifier (configured for some closed-loop gain that is TBD) that tolerates an open input. Because of our experience with the LM4562, I am inclined to keep each stage in a different chip so that physical separation can be achieved; that is, we can keep the scope probe input as well separated as possible from the output of the 2nd stage. Separation is difficult to achieve in a single IC package. It may be that simply preceding the LM4562 with a simple unity gain buffer would be sufficient as that would allow driving the LM4562 from a low impedance.

I had considered building a DIY differential scope probe some years ago, but after a bit of study I decided that I had little chance of achieving wide differential bandwidth (at least 10MHz) and high common mode rejection (at high frequency); that's one reason why commercial differential probes are so expensive--they are not easy to design. Also, beyond a few exploratory tests, each design iteration would require fabricating a new PCB; breadboarding would not suffice. A dozen or two PCB iterations could get expensive and I have neither skills nor experience in RF design.

You could also consider whether requiring the LM4562 to be stable with an open input is reasonable. In actual use, surely you would always have a probe connected.

@Dave Lowther
Some odds & ends...
I do think an experiment would be useful to determine how low the gain of the 4562 stage must be made for stable/no-oscillation operation (as you suggest in post#63). That info could help determine a 2-amp approach.
Splitting the gain of 100 between two stages can have advantages; stability is improved, bandwidth of each stage (but not necessarily after both stages in series) is greater. If the gain of the first stage is high enough (typically not very high,5-15), then the noise contribution of the second stage disappears and noise is set by the first stage. If the first stage has only unity gain, then its noise adds to the noise of the second stage. Thus, the first stage ought to be quiet, fast, have low bias current (for high resistor to ground). The 4562 is weakened only by its high bias currents.

 

@TeeKay6

You have a compulsion to neatness that is not necessary for tests, since they can be disassembled as soon as the test is complete. However, in general, neatness alone will not invalidate a test. And you do have a very neat "paper trail" of your work. However, neatness does not in general lead to better circuit performance.

You are right, I am compelled to try and be neat. If you are specifically talking about the added 2x100nF+1M if it's neat then it's only because I was trying to keep the leads as short as possible.

I will accept that you have met the requirement of a single ground point with all unnecessary components disconnected from the amp
I have not traced out your test wiring; I will accept that you have met the requirement of a single ground point with all unnecessary components disconnected from the amp.

I still need to do the change to R4,C5 as I mentioned in comment #65.

Your stated result is that the amp still oscillates and I assume that means the waveform is still a rough squarewave--i.e. oscillation is not marginal but quite dominant.

Yes it looks like it did before the change except for higher frequency and the slopes of the edges are more obvious.

You could also consider whether requiring the LM4562 to be stable with an open input is reasonable. In actual use, surely you would always have a probe connected.

I wouldn't like to own a scope that drew a square wave on the screen if no probe was connected. I do take your point though, it's pragmatic.

I do think an experiment would be useful to determine how low the gain of the 4562 stage must be made for stable/no-oscillation operation (as you suggest in post#63). That info could help determine a 2-amp approach.

I'll do that experiment.
Thanks again for your advice.
Dave

 

@TeeKay6
You are right, I am compelled to try and be neat. If you are specifically talking about the added 2x100nF+1M if it's neat then it's only because I was trying to keep the leads as short as possible.
I still need to do the change to R4,C5 as I mentioned in comment #65.
Yes it looks like it did before the change except for higher frequency and the slopes of the edges are more obvious.
I wouldn't like to own a scope that drew a square wave on the screen if no probe was connected. I do take your point though, it's pragmatic.
I'll do that experiment.
Thanks again for your advice.
Dave

@Dave Lowther
If a man's worst defect were being neat, then surely our world would be much better than it is.

 

@TeeKay6
I did the change to R4,C5 as I mentioned in comment #65. It made no difference.
I've removed the temporary components I added and repaired the tracks I cut.
So now it's back to fig37 state (decoupling modified, but no mimimalisation).
With it back in the die cast box, it oscillates as before with nothing connected to the BNC input.
With a scope probe connected to the BNC input it doesn't oscillate on x1 or x10 scope probe switch setting.
The DC resistance of the scope probe measured from its BNC connecter is open circuit
The capacitance of the scope probe measured from its BNC connector is about 135 pF measured using relative C range on my multimeter.
I've still got to do the experiment to see what value of R across JP1 will make it stable at various gains.

 

a side remark - it's those 2MΩ thingies -- the 2.2MΩ is the resistance (usually) above what (more precisely below 300nA) too few electrons are involved in a current flow e.g. where fields start to "dial in" (with sensitive / high gain amps the circuit (also simple longer wires of it - ) pick(s) up mechanical tapping to the table and changes in the ais'r electrical charges / grounding capacitors -- such as human) ??? ( a - self-check : ) https://sciencing.com/convert-amps-electrons-per-second-8628812.html → "1 ampere = 6.242·10¹⁸ electrons per second" · 3E-7 = 1.8726·10¹² . . . ? perhaps a couple of more would help the situation ???

 

a side remark - it's those 2MΩ thingies -- the 2.2MΩ is the resistance (usually) above what (more precisely below 300nA) too few electrons are involved in a current flow e.g. where fields start to "dial in" (with sensitive / high gain amps) ??? ( a - self-check : ) https://sciencing.com/convert-amps-electrons-per-second-8628812.html → "1 ampere = 6.242·10¹⁸ electrons per second" · 3E-7 = 1.8726·10¹² . . . ? perhaps a couple of more would help the situation ???

Thanks for the comment ci139. I'm trying to keep the input impedance at >= 1MΩ so that it is similar to a scope. I just found out that if I put a 2MΩ pot across the input (JP1), the oscillation starts when I turn the pot up from 0Ω to 50kΩ. Similarly a capacitor of 82pF across JP1 is just enough to prevent oscilation, 68pF is not enough. As I said in recent comment, I'm going to experiment to find out what reduced gain stabilises it and prevents the oscillation when JP1 is open circuit.
Best regards
Dave

 

when i made a mic. amp i soldered everything to the other side of the 8-DIP socket's leads -- and it didn't pick up anything but the electro-dynamic in series carbon input /// while when testing (i don't remember anymore what) in 3D open air wire connectors - the mentioned effects poped in (when the amp was near ideally biased) . . .

 

when i made a mic. amp i soldered everything to the other side of the 8-DIP socket's leads -- and it didn't pick up anything but the electro-dynamic in series carbon input /// while when testing (i don't remember anymore what) in 3D open air wire connectors - the mentioned effects poped in (when the amp was near ideally biased) . . .

I ordered some copper clad board today so that I could build it in 'dead bug' style should I get the urge.

 

. . . R5:R4 10k:100Ω . . .
. . . in simulations : most common is cancelling ripple/instability/osc by shunting R5 with 500fF to 5pF (← this applies only when using 1 capacitor "integrator" . . . there are also exceptions to this magnitude)
--or-- bridging both with (inverse-/reciprocal-) proportional 1-s ( --or(further)-- both shoulders of the op amp)
--and--
less frequently (actually rarely e.g. in "specific situations" ) grounding the +inp with "cancelling / balancing" capacitor for the "capacitive biasing system"
(the least been used in conjunction with experimental ' "abs." voltage refferences' e.g. high gains --and/or-- short response times)
___________
the local resource → :: https://www.allaboutcircuits.com/te...feedback-part-7-frequency-dependent-feedback/ (may require logout or opening in different browser (you're not logged in with))
____________
the LM4562 has 20x lower input current than LM318 and is a way faster . . . LM318 has a better variety of app-notes however http://www.ti.com/product/LM318/technicaldocuments
(for example the pg.25 and on)

 

. . . R5:R4 10k:100Ω . . .
. . . in simulations : most common is cancelling ripple/instability/osc by shunting R5 with 500fF to 5pF (← this applies only when using 1 capacitor "integrator" . . . there are also exceptions to this magnitude)
--or-- bridging both with (inverse-/reciprocal-) proportional 1-s ( --or(further)-- both shoulders of the op amp)
--and--
less frequently (actually rarely e.g. in "specific situations" ) grounding the +inp with "cancelling / balancing" capacitor for the "capacitive biasing system"
(the least been used in conjunction with experimental ' "abs." voltage refferences' e.g. high gains --and/or-- short response times)
___________
the local resource → :: https://www.allaboutcircuits.com/te...feedback-part-7-frequency-dependent-feedback/ (may require logout or opening in different browser (you're not logged in with))
____________
the LM4562 has 20x lower input current than LM318 and is a way faster . . . LM318 has a better variety of app-notes however http://www.ti.com/product/LM318/technicaldocuments
(for example the pg.25 and on)

@Dave Lowther
Because you are aiming for the highest bandwidth I have not suggested certain approaches to stabilizing the 4562. However, if only for info purposes, it may be worthwhile to try (as @ci139 notes) a tiny bit of compensation in the form of a few pF cap across the feedback resistor. Start very small (1pF, 3pF, 5pF, 10pF...) and increase until circuit is stable or until the cap is obviously too large. As you try values, you look for stability and effect on bandwidth (hi freq response). Continue to keep your meticulous records as you experiment so that you can later review & decide what is the best next action. If you can achieve stability with a relatively small feedback cap, then placing a resistance in series with that cap would be a logical next step to regain some bandwidth. Short leads at the opamp input are critical, as is separation of output signals from input signals. There are other schemes for compensation but they can become quite complex; also, you may find that what works changes with only a small change to breadboard layout. I find it discouraging that even with nothing more than a short lead to a high value resistor hanging on the + input, oscillation remains strong. However, one useful rule is: when nothing works, try something else...any something else. With luck you may stumble upon what you need.

 

In reply to comment #75 and comment #76.
@ci139 and @TeeKay6
Thanks for your advice on stabilisation.
My reason for using the LM4562 is its spec for THD and noise which is described in the data sheet (top right of front sheet) as "extremely low voltage noise density (2.7nV/√Hz) with vanishingly low THD+N (0.00003%)".
My reason for being concerend about THD and noise is that I want to use this front end amplifier with a Roland Rubix 22 sound 'card' to measure THD and noise of any amplifier circuit I may build in the future.
My bandwidth requirement is at least 20 kHz and ideally 80 kHz to match the Roland when it's sampling at 192 kHz (1dB down at 60kHz, 5dB down at 80 kHz). In theory that would let me measure THD at 20 kHz up to 3rd harmonic and see some of the 4th Harmonic. The LM4562 min GBWP is 45 MHz. Gain 100 x 20 kHz is 2MHz. I'm not sure how much affect getting that 'close' to GBWP will have on the THD at 20 kHz. Presumably implementing the gain in two stages each with a gain of 10 for example would help reduce THD at 20 kHz.
This morning I have removed R4 and replaced it with header pins so that I can easily fit different values of R4 for testing. Next I'm intending to investigate stability for various gains (various values of R4). I'll report back when I've done that.
Best regards
Dave.

 

In reply to comment #75 and comment #76.
@ci139 and @TeeKay6
Next I'm intending to investigate stability for various gains (various values of R4). I'll report back when I've done that.
Best regards
Dave.

I've done that and the results are in the table:

The 'Required Gain' column contains the switchable gains that I intend to implement in the final build. "ST" means "Straight Through", i.e.the signal goes in and out of the die cast box with the board in it, but the signal is routed directly from the BNC Input to the BNC output without going to the board. "Direct" is the signal generator connected directly to the sound card bypassing the die cast box completely.
The "Osc O/C" column indicates whether it oscillates when there is nothing connected to the BNC input on the die cast box. If it doesn't oscillate I've recorded the RMS voltage measured at the MI oscilloscope software. If it does oscillate I've indicated the frequency it oscillates at when the short length of screened cable from the input BNC socket is connected to JP1. With literally nothing connected to JP1 the oscillation frequency is higher at about 30 kHz.
As the stability and performance looks OK up to a gain of 20, I should be able to proceed with a two stage solution where the first stage gain is limited to 10 or 20. The max gain of the 2nd stage will need to be 10 or 5.
I'm going to have to take a break from working on this as I've been putting off starting some house DIY until I got this amplifier issue to a point where I could park it for a while without spending all my time wondering why it was oscillating and how to fix it.
Thanks again to everyone who has helped me here.
Best regards
Dave

 

using the LM4562

to be honest - i myself wouldn't consider using even the LM318 unless there is no other options left

you boldly go forward with the one of the best amplifiers on the Planet (http://www.ti.com/lit/sg/slot140/slot140.pdf)

it is why this reply - to note that out

 

I've done that and the results are in the table:
View attachment 185111
The 'Required Gain' column contains the switchable gains that I intend to implement in the final build. "ST" means "Straight Through", i.e.the signal goes in and out of the die cast box with the board in it, but the signal is routed directly from the BNC Input to the BNC output without going to the board. "Direct" is the signal generator connected directly to the sound card bypassing the die cast box completely.
The "Osc O/C" column indicates whether it oscillates when there is nothing connected to the BNC input on the die cast box. If it doesn't oscillate I've recorded the RMS voltage measured at the MI oscilloscope software. If it does oscillate I've indicated the frequency it oscillates at when the short length of screened cable from the input BNC socket is connected to JP1. With literally nothing connected to JP1 the oscillation frequency is higher at about 30 kHz.
As the stability and performance looks OK up to a gain of 20, I should be able to proceed with a two stage solution where the first stage gain is limited to 10 or 20. The max gain of the 2nd stage will need to be 10 or 5.
I'm going to have to take a break from working on this as I've been putting off starting some house DIY until I got this amplifier issue to a point where I could park it for a while without spending all my time wondering why it was oscillating and how to fix it.
Thanks again to everyone who has helped me here.
Best regards
Dave

@Dave Lowther
Beautiful work, Dave! If indeed you ultimately decide on a 2-stage approach, there ought to be many candidates (incl LM4562) for the 2nd stage that will not unduly degrade the performance of the first stage. Certainly a break from your intense concentration on this project is well earned. I will still wonder if (in the future) trying the gain=50 condition ought to be tried with the addition of a very small cap (a few pF) paralleling the 10K feedback R to learn if that helps or hinders. Good luck with your other DIY home projects!