I took the advice of JT and others to heart and scrapped my original design, and spent some time working this one out with pencil and paper before approaching the simulator. I think the following design is quite a bit improved, considering at first take it actually appears to be stable!
This one is DC coupled with a servo to keep the output centered. The output stage is class A and I think should drive a pair of 64 ohm headphones to ear bleeding levels without a problem.
Schematic:
formatting link
Frequency and phase response:
formatting link
FFT/distortion:
formatting link
(I think the stuff at around 100 kHz and above is an artifact of the FFT as it also appears when I do an FFT of a zero impedance sine source by itself.)
Comments and/or criticism would be welcome and appreciated.
Still too many parts... AND it's oscillating ;-) ...Jim Thompson
--
| James E.Thompson, CTO | mens |
| Analog Innovations, Inc. | et |
| Analog/Mixed-Signal ASIC's and Discrete Systems | manus |
| Phoenix, Arizona 85048 Skype: Contacts Only | |
| Voice:(480)460-2350 Fax: Available upon request | Brass Rat |
| E-mail Icon at http://www.analog-innovations.com | 1962 |
Remember: Once you go over the hill, you pick up speed
Too much spectral activity at the high end smells of birdies. ...Jim Thompson
--
| James E.Thompson, CTO | mens |
| Analog Innovations, Inc. | et |
| Analog/Mixed-Signal ASIC's and Discrete Systems | manus |
| Phoenix, Arizona 85048 Skype: Contacts Only | |
| Voice:(480)460-2350 Fax: Available upon request | Brass Rat |
| E-mail Icon at http://www.analog-innovations.com | 1962 |
Remember: Once you go over the hill, you pick up speed
I was confused by that too, but when I ran the FFT on just the output sine wave source connected to nothing the picture looked the same except with the 2nd harmonic spike gone. All those spikes above 100 kHz were still there on what should be a pure sine wave. That's why I think it's an artifact of the FFT... I haven't been able to get FFTs to measure distortion in LTSpice to work consistently.
with a crude OpAmp, though I don't know why you don't just use my first posted version with a real OpAmp.
To meet performance equivalent to using a real OpAmp will require many more parts to accomplish current sources, but you'll never meet the matching attainable in a real OpAmp. ...Jim Thompson
--
| James E.Thompson, CTO | mens |
| Analog Innovations, Inc. | et |
| Analog/Mixed-Signal ASIC's and Discrete Systems | manus |
| Phoenix, Arizona 85048 Skype: Contacts Only | |
| Voice:(480)460-2350 Fax: Available upon request | Brass Rat |
| E-mail Icon at http://www.analog-innovations.com | 1962 |
Remember: Once you go over the hill, you pick up speed
Thanks Jim, I will give that circuit a try as well, though I have constructed amps with a similar topology before aside from the output stage. Part of the reason in doing this project is for the satisfaction of being able to say "this is an original design of mine," even if the final circuit might not be the most optimal or part parsimonious! :)
Pessimistic :( That's OK, I appreciate your candor. Will there really be a thermal runaway hazard at such relatively low currents in Q5 and Q6? I don't think they ever see much more current than a mA or two. I suppose some emitter resistors couldn't hurt.
One of those rare occasions when Larkin is correct. When Q5 and Q6 heat up, their VBE's DECREASE relative to Q2/Q3, thus they draw more current, heat up more...
I doubt if distortion will be bad... until it runs away ;-)
You still have too many parts.
Pretend you're trying to sell your boss on the need for each transistor... write down a description of the need for each.
I'd like to see an overall description of circuit function.
Writing it down might cause you to eliminate some parts ;-)
BTW: Q8, Q9 and Q10 look like an open invitation to a "songfest" ;-) ...Jim Thompson
--
| James E.Thompson, CTO | mens |
| Analog Innovations, Inc. | et |
| Analog/Mixed-Signal ASIC's and Discrete Systems | manus |
| Phoenix, Arizona 85048 Skype: Contacts Only | |
| Voice:(480)460-2350 Fax: Available upon request | Brass Rat |
| E-mail Icon at http://www.analog-innovations.com | 1962 |
Remember: Once you go over the hill, you pick up speed
Your first gain stage is single ended. That generally doesn't produce a low distortion design.
Nothing wrong with rolling your own design as long as you are learning in the process.
Incidentally, the FFT in spice requires a fixed time step to be remotely accurate. It is the 4th parameter on the .tran line in most spice implementations.Spice, if left to it's own "thinking", picks a time step based on accuracy parameters, then interpolates to snap on your time grid. With the 4th parameter, spice is forced to converge on a solution at each time step. It runs much slower, but that hardly matters these days.
I'd still suggest going MOS on the output stage. Bipolar emitter followers can often have surprising results in real life. They essentially look inductive if you think about it. You can get some real life high frequency oscillations.
You should do some step response simulations. It is best to to a large step to check slewing, and a small step to check stability. Ideally positive and negative steps will be symmetric.
I guarantee I can build a better (audio) power amplifier with bipolar followers, rather than a MOS output.... class AB AND thermally stable :-) ...Jim Thompson
--
| James E.Thompson, CTO | mens |
| Analog Innovations, Inc. | et |
| Analog/Mixed-Signal ASIC's and Discrete Systems | manus |
| Phoenix, Arizona 85048 Skype: Contacts Only | |
| Voice:(480)460-2350 Fax: Available upon request | Brass Rat |
| E-mail Icon at http://www.analog-innovations.com | 1962 |
Remember: Once you go over the hill, you pick up speed
One of the advantages of doing this "just as a hobby" is that I don't need to engage in that sort of bean counting. I'm free to try out things just for the sake of being different or for the joy of making something work that hasn't been done before. Kind of like the guys who build 8 bit computers out of TTL logic - completely pointless from a practicality point of view as one could buy an 8 bit microcontroller with ten times the processing power. One might say that it's the process that's more rewarding than the end result, and hopefully one learns something along the way.
If I should ever attempt to do circuit design as a living, as you clearly point out this attitude will have to shift. Clearly, a design that uses say, a four-pole filter when a two-pole will do is a losing design from an economic standpoint. I have other projects that I'm working on where I consciously attempt to minimize the complexity and maximize the performance.
Of course, the downside to this method is that there is always a net flow of dollars in the wrong direction...:(
As I see it, it essentially has 3 stages: a cascode class A gain stage, a buffer, and a class A output stage. The buffer also serves the purpose of providing the boostrapping to give the input stage its large gain, which is tamed down somewhat by local feedback through the emitter degeneration resistor. My hope is that this will provide an input gain stage with good bandwidth, noise, and PSRR, and relatively low distortion (not as low as a differential pair as has been pointed out, but 2nd harmonic is considered to be "pleasing").
The buffer is also arranged so that the combination of Vbe rises and falls allows both the input stage and the output to sit at 0 volts, allowing the maximum headroom for the bootstrapped gain stage, and also allowing the output to sit at zero volts for DC coupling. The bias servo should make it stay close to ground to keep DC current out of the headphones.
The output stage is a variation on a circuit I saw elsewhere; I guess you might call it push-pull class A. R14 is sized so that at the quiescent point it carries approximately the same current as the peak current into the load for the desired output power. As Q8 turns on the increased drop will start to shut down Q10, and the reduction in current through Q10 will be replaced by current from the load. As Q8 turns off the decrease in voltage across R14 will cause Q10 to source current into the load. At no point in the cycle does either transistor go into cutoff, though. The diode prevents Q9 from saturating and having its base current take off on the negative cycle.
A possible flaw I do see with this arrangement is that while Q8 has its base capacitance actively charged and discharged by the buffer, Q10 has to have its base capacitance charged up by R11, so maybe it'll be "sluggish" compared to its counterpart. Perhaps this won't be a problem though because as I was surprised to find out the KSA1173 has a Ft of
100 mHz. I don't know if they're any good for audio, though - Mouser was blowing out the non-RoHS KSC1173 and its PNP counterpart the KSA473 a while back for about 3 cents each so that's why I have some. :)
I'll have to breadboard it and see! I'll probably have to order some 2 watt power resistors for R14. Thanks again for your suggestions and taking the time to comment.
Thanks for your reply - I'll take your advice and do the step response tests; what exactly am I looking for output aside from symmetry? I don't have a lot of experience designing with power MOSFETS aside from some experiments with boost converters, but I could certainly make an attempt at designing an output stage that uses them. I don't think the the present output stage would directly translate.
My suggestion was to get you to think about function... not specifically bean counting.
[snip]
3 stages, with overall feedback, is ALWAYS an open invitation to instability.
IF you only think voltage gain.
Why don't you analyze my suggestion, just for yucks ?:-) ...Jim Thompson
--
| James E.Thompson, CTO | mens |
| Analog Innovations, Inc. | et |
| Analog/Mixed-Signal ASIC's and Discrete Systems | manus |
| Phoenix, Arizona 85048 Skype: Contacts Only | |
| Voice:(480)460-2350 Fax: Available upon request | Brass Rat |
| E-mail Icon at http://www.analog-innovations.com | 1962 |
Remember: Once you go over the hill, you pick up speed
Should be done for both pull-up and pull-down conditions due to class-B gain asymmetries.
...Jim Thompson
--
| James E.Thompson, CTO | mens |
| Analog Innovations, Inc. | et |
| Analog/Mixed-Signal ASIC's and Discrete Systems | manus |
| Phoenix, Arizona 85048 Skype: Contacts Only | |
| Voice:(480)460-2350 Fax: Available upon request | Brass Rat |
| E-mail Icon at http://www.analog-innovations.com | 1962 |
Remember: Once you go over the hill, you pick up speed
Amongst folks who do this for a living, part of the game is to minimize parts count, not just to shave pennies, but because it's part of the game. I've seen a lot of audio designs where the author seemed to want to maximize parts count.
The only danger with too few parts is having one part perform more than one function, which can result in the "not enough knobs to turn" dilemma.
You could make a good discrete headphone amp with about a dozen parts, less if you don't mind a little dissipation.
Bitrex claimed the need for 8V P-P into 64 Ohms... 62.5mA Peak.
I showed a simple minded buffer plus an OpAmp, but he's trying to learn to "design", without any math or semiconductor background ;-) ...Jim Thompson
--
| James E.Thompson, CTO | mens |
| Analog Innovations, Inc. | et |
| Analog/Mixed-Signal ASIC's and Discrete Systems | manus |
| Phoenix, Arizona 85048 Skype: Contacts Only | |
| Voice:(480)460-2350 Fax: Available upon request | Brass Rat |
| E-mail Icon at http://www.analog-innovations.com | 1962 |
Remember: Once you go over the hill, you pick up speed
ElectronDepot website is not affiliated with any of the manufacturers or service providers discussed here.
All logos and trade names are the property of their respective owners.