Hi kt, have you tried a loop gain analysis on the amp yet? There's an example in the LTSpiceIV/examples/Educational/LoopGain2.asc on how to do it. It would be best to add parasitic properties to the circuit: small inductors/resistors on each node to simulate the interconnecting wires (2-20nH, 1-5mOhm for instance), depending on how long the wires seem to be in the real circuit. Also, always add ESR and ESL to all capacitors. I usually add around 15nH ESL to each capacitor; for electrolytic caps the ESR varies widely, but something between 0R1 and 0R5 is not far from reality. For film caps I usually add ESR in the tens of milliohms, such as 50mR, or 70mR, or 20mR. These ballpark values come from measuring several components.
I posted the open-loop plot in #174. I've been using open loop gain stuff to tune stability in the simulator for some time. Admittedly, knowledge of stability can determine whether a design lands in the graveyard or in real implementation...
Caps critical to stability (C5 and C2) were simulated with at least .1 ohms, but earlier I was using 10 ohms each without problems. Series resistance of at least 10R is actually necessary for C5, to lower the Q. I don't know if it is possible to have such a steep OLG curve without Q-limiting at least one stability cap. I just simulated with 10nH of series inductance and stability changed little. So it seems that, if that series resistor works, this amp will be more stable than we expected.
Here is the new plot with parasitics added to all caps.
- keantoken
Caps critical to stability (C5 and C2) were simulated with at least .1 ohms, but earlier I was using 10 ohms each without problems. Series resistance of at least 10R is actually necessary for C5, to lower the Q. I don't know if it is possible to have such a steep OLG curve without Q-limiting at least one stability cap. I just simulated with 10nH of series inductance and stability changed little. So it seems that, if that series resistor works, this amp will be more stable than we expected.
Here is the new plot with parasitics added to all caps.
- keantoken
Attachments
Last edited:
What I MEANT to say in post #180, what that LOWERING C5's series R to .1 ohms resulted in oscillation. However, revert to 10R series resistance and it stabilizes, and produces the OLG plot I just posted. I was originally simulating with 10R series resistance... So I have an idea of why it's oscillating. I need to add the resistor.
I already have a plan for what to do next, just can't do it right now.
- keantoken
I already have a plan for what to do next, just can't do it right now.
- keantoken
Very tough to add inductors on wires
Gah, I'm tired already! What can I say...
Well you must know it's people like you who come in and tell me all about parasitics and get me paranoid, and then I go and try to connect components together with 1mm of clearance
(although to my way of thinking it's better to work with the component and try not to splay pins...). Probably not a bad thing though, and I'm getting better.Here is the frequency doubler, amplified since the input/output volume ratio is a bit low... Now you get in almost as much as you get out. FFT plot of the amplified circuit, showing just what is possible with this trivial configuration... Matching doesn't matter, just tune out the fundamental with the bias trimmer. .asc file included (rename to .asc).
I'm printing off a bunch of schematics such as this one and am going to go to a HAM radio club meeting, maybe scrape up some equipment.
- keantoken
Attachments
Gah, I'm tired already! What can I say...
Well you must know it's people like you who come in and tell me all about parasitics and get me paranoid, and then I go and try to connect components together with 1mm of clearance
- keantoken
Sorry man, just an opinion. Don't mean to curb your enthusiasm. I was talking about the ltspice schematic, your style of drawing components very close to each other. For me that makes it very hard to a) add inductors on wires between components (trying to emulate the real circuit) and b) to see the bigger picture. But I edited the schematic, lots of space now (one needs a large screen for this).
Have fun at the meeting.
I understand your concern, but I'm doing something else right now. Enthusiasm isn't the variable... Maybe you misinterpreted my half-baked attempt at humor...
I honestly didn't think of adding series inductance on the schematic, though I am now well aware of series R.
All the wires are very short, practically the same length. 22nH on every lead should do it justice...
- keantoken
I honestly didn't think of adding series inductance on the schematic, though I am now well aware of series R.
All the wires are very short, practically the same length. 22nH on every lead should do it justice...
- keantoken
Last edited:
It's working!
See Iko, my construction is more than adequate. I don't think PCB's or protoboards are really that necessary, at least for a simple design like this. (for prototyping yes, the PCB is still in the works)
It is stable, just tuned the bias pot, it's showing a 3.5V pk-pk sine from my signal generator.
Here is the schematic, as built (inductors excluded). The compensation is large because I only have 22nF caps. All it needed was 10R in series with C5, EXACTLY as the simulator predicted. I've also attached the open loop plot with the inductors.
Checking with the signal generator, there is no peaking in the response.
The HAM meet was almost interesting... One of the guys was impressed with the frequency doubler, and told me if it works well I could write an article for the local HAM newsletter... Said it would be useful in getting a clean 10MHz frequency standard from the broadcasted 5MHz standard. But I haven't tested it at RF, and I would probably choose different transistors...
- keantoken
See Iko, my construction is more than adequate.
I don't think PCB's or protoboards are really that necessary, at least for a simple design like this. (for prototyping yes, the PCB is still in the works)It is stable, just tuned the bias pot, it's showing a 3.5V pk-pk sine from my signal generator.
Here is the schematic, as built (inductors excluded). The compensation is large because I only have 22nF caps. All it needed was 10R in series with C5, EXACTLY as the simulator predicted. I've also attached the open loop plot with the inductors.
Checking with the signal generator, there is no peaking in the response.
The HAM meet was almost interesting... One of the guys was impressed with the frequency doubler, and told me if it works well I could write an article for the local HAM newsletter... Said it would be useful in getting a clean 10MHz frequency standard from the broadcasted 5MHz standard. But I haven't tested it at RF, and I would probably choose different transistors...
- keantoken
Attachments
Well, I'm doing some listening.
It is somewhat unusual since it's stereo and I'm listening to mono. The lack of stereo separation causes fatigue... However the bass sounds really good, and all the instruments are separated by a lot of air. There's no melodic soup in the background to slur the notes together...
Okay, I just heard some background stage props! An initiatory experience! Wow!
Well, the circuit works, and if I can get it stable in air sculpture then I sure as hell can get a PCB working... Then I can test out stereo. I think turning up the niceness knob makes it sound slightly "brighter". Not sure, I can barely tell...
- keantoken
It is somewhat unusual since it's stereo and I'm listening to mono. The lack of stereo separation causes fatigue... However the bass sounds really good, and all the instruments are separated by a lot of air. There's no melodic soup in the background to slur the notes together...
Okay, I just heard some background stage props! An initiatory experience! Wow!
Well, the circuit works, and if I can get it stable in air sculpture then I sure as hell can get a PCB working... Then I can test out stereo. I think turning up the niceness knob makes it sound slightly "brighter". Not sure, I can barely tell...
- keantoken
Once I get the PCB done I could send you a board. Trade for an IkoReg, perhaps?
Listening to Steely Dan, Reelin' in the years, I can discern the instruments behind the vocals and distorted guitar...
It sounds NATURAL!!! Well, there are some things about the sound I'm not sure I like, but this is so different I'm not sure if I'm just overreacting. In any case, the compensation is much larger than it needs to be, I suspect this might have an effect. I need to order some good caps in the near future...
- keantoken
Listening to Steely Dan, Reelin' in the years, I can discern the instruments behind the vocals and distorted guitar...
It sounds NATURAL!!! Well, there are some things about the sound I'm not sure I like, but this is so different I'm not sure if I'm just overreacting. In any case, the compensation is much larger than it needs to be, I suspect this might have an effect. I need to order some good caps in the near future...
- keantoken
Last edited:
Member
Joined 2009
Paid Member
very nice work Keantoken !
You know, almost all of the components we use are non-ideal and you could define a subckt for them so as to avoid cluttering up your schematic with all those parasitics. But I'm not so sure it's worth the effort ?
The niceness knob has a very subtle effect. I discovered this too with the 'TGM goes Tubey' design. I have to add a lot of harmonic distortion to get an obvious result. It turns out that it is more obvious when I listen for an extended time. In my case I found that my subconsious picks up on it and I find that I simply prefer listening to one amplifier (or version thereof) more than another without being able to put a finger on the exact reason why..... and Hugh's AKSA (which has it's own subtleties) is a good example.
The HAM radio guys are much better versed in designing around parasitics than most audio DIYers. The manhatten construction technique for example would be a great way to build a pre-amp.
You know, almost all of the components we use are non-ideal and you could define a subckt for them so as to avoid cluttering up your schematic with all those parasitics. But I'm not so sure it's worth the effort ?
The niceness knob has a very subtle effect. I discovered this too with the 'TGM goes Tubey' design. I have to add a lot of harmonic distortion to get an obvious result. It turns out that it is more obvious when I listen for an extended time. In my case I found that my subconsious picks up on it and I find that I simply prefer listening to one amplifier (or version thereof) more than another without being able to put a finger on the exact reason why..... and Hugh's AKSA (which has it's own subtleties) is a good example.
The HAM radio guys are much better versed in designing around parasitics than most audio DIYers. The manhatten construction technique for example would be a great way to build a pre-amp.
First of all, I don't think simulation of parasitics can go a long way, especially if you can build the thing before you get done setting up the simulation. It is important to throw in some inductors, and see the results. In a few unfortunate circuits, even a tiny inductor in some spot will throw the OLG into convulsions, and thus they won't be workable. Put small inductors into critical paths, and see what happens.
I figured out what those caps were. They are X7R type ceramics, 22nF +-10%. Think it's okay to use film caps in those locations? I suppose I'll have to try. I got a bunch of film caps from a broken amp, for audio use, but series resistance doesn't matter much for the compensation here.
Just replaced C2 with a 10n MKT... I'll probably go no lower than 6.8n, then I'll try different things with C5.
This biggest difference (and it is NOT subtle!) I hear between this amp and others is in the lowest 3 octaves. I don't know why the difference, but it is obvious. What is strange is that one of the synthesized instruments I made, actually sounds MORE real.
- keantoken
I figured out what those caps were. They are X7R type ceramics, 22nF +-10%. Think it's okay to use film caps in those locations? I suppose I'll have to try. I got a bunch of film caps from a broken amp, for audio use, but series resistance doesn't matter much for the compensation here.
Just replaced C2 with a 10n MKT... I'll probably go no lower than 6.8n, then I'll try different things with C5.
This biggest difference (and it is NOT subtle!) I hear between this amp and others is in the lowest 3 octaves. I don't know why the difference, but it is obvious. What is strange is that one of the synthesized instruments I made, actually sounds MORE real.
- keantoken
Last edited:
I can't get to that Manhattan PDF, since the link is down now, I should have saved it.
I've been trying to create a virtual earth, so I can DC block the output for protection, but it seems there is no way to do this without either losing OLG at low frequencies, or having a capacitor in the output signal path. So I have to digress; I recommended Bigun that a passive virtual earth would protect speakers without a cap in the signal path but it turns out this is not true. In order to have a clean signal path to the speaker, input ground must be connected to power ground, preferably with no more than a resistor and what other components you consider inert. This is because the amp must compare the input and feedback against the same ground; if this reference is split, the difference will be added to the output voltage.
I don't think there is much more I can tell unless I get a stereo pair...
- keantoken
I've been trying to create a virtual earth, so I can DC block the output for protection, but it seems there is no way to do this without either losing OLG at low frequencies, or having a capacitor in the output signal path. So I have to digress; I recommended Bigun that a passive virtual earth would protect speakers without a cap in the signal path but it turns out this is not true. In order to have a clean signal path to the speaker, input ground must be connected to power ground, preferably with no more than a resistor and what other components you consider inert. This is because the amp must compare the input and feedback against the same ground; if this reference is split, the difference will be added to the output voltage.
I don't think there is much more I can tell unless I get a stereo pair...
- keantoken
kt, there's nothing complicated about the Manhattan island style. You start with a piece of blank pcb and a number of "islands" which are just small pieces of pcb. They get either soldered or glued to the surface of larger pcb. These islands are the nodes of your circuit, where all the components that get connected to each other get soldered. Some people use a puncher to get round islands. I just used a saw to cut square/rectangular little pieces. That's all there is to it. These days for my prototypes I just use thicker copper wire instead of these islands, and sometimes a blank piece of pcb as the ground plane.
One thing is odd. The circuit doesn't clip like the simulator predicts. Clipping actually seems healthier than in the sim.
Doing further tests, I'm certainly considering switching to BJT instead of MOSFET in the output stage. The high gate capacitance causes a lot of trouble as well as decreasing slew rate and really bad clipping behavior.
I replaced C5 with a 10n MKT, seemed to sound better but something has been causing spikes at the input of my amp, so I don't feel like endangering my headphones. Curses!
- keantoken
Doing further tests, I'm certainly considering switching to BJT instead of MOSFET in the output stage. The high gate capacitance causes a lot of trouble as well as decreasing slew rate and really bad clipping behavior.
I replaced C5 with a 10n MKT, seemed to sound better but something has been causing spikes at the input of my amp, so I don't feel like endangering my headphones. Curses!
- keantoken
Last edited:
The MOSFET problem is possibly related to low drain voltage?
Especially when VDS<VGS.
I've been warned by several high muckimucks round here that
Miller capacitance rises sharply as drain voltage falls. But you
rarely ever see severity of this non-linear effect hinted at in
spec sheets, and never in a sim model.
This is just my own unconfirmed theory hereafter:
You think about it, MOSFET is built fairly symmetrical inside.
Why should Cgd ever measure less than Cgs in first place?
Maybe without bias voltage, those two figures converge.
Especially when VDS<VGS.
I've been warned by several high muckimucks round here that
Miller capacitance rises sharply as drain voltage falls. But you
rarely ever see severity of this non-linear effect hinted at in
spec sheets, and never in a sim model.
This is just my own unconfirmed theory hereafter:
You think about it, MOSFET is built fairly symmetrical inside.
Why should Cgd ever measure less than Cgs in first place?
Maybe without bias voltage, those two figures converge.
Last edited:
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.