I thought dropping the impedance of the Class A side might help
to quench some of the ripple? but I can't tell any difference one
way or the other....
Adding split sense on the left works fine. But if you tie the load
also splitting the main sense in the middle, goes straight to hell
in-a-handbasket right quick like. I don't know why...
The series 2K2's were for a bootstrap cap, but the sim takes far
too many cycles for that to reach equilibrium. Wasn't needed for
the input signal level as given, so I ditched it.
I failed to improve anything
to quench some of the ripple? but I can't tell any difference one
way or the other....
Adding split sense on the left works fine. But if you tie the load
also splitting the main sense in the middle, goes straight to hell
in-a-handbasket right quick like. I don't know why...
The series 2K2's were for a bootstrap cap, but the sim takes far
too many cycles for that to reach equilibrium. Wasn't needed for
the input signal level as given, so I ditched it.
I failed to improve anything
I like how you re-jigged that to abuse the sense resistor again Ken 
I think the ripple results from lots of different factors in this circuit. Probably the hysteresis in the comparator is the biggest factor. Increasing the value of the inductor slows the rate of change in current injected by the switches, so for the same switch rate the ripple magnitude goes down with a bigger inductor. Of course this also reduces the slew rate into the load - which is where I think the linear amp was being called upon to fill the gap in that thesis paper. At this stage I was trying to keep the simulation minimal for the sake of clarity.
A differential amp before the comparator should help free-up some of the parameters we're working with here. I was a bit constrained by the 0-5V range of the comparator inputs, and the resistive divider was an all-too-convenient solution really.
I'd like to breadboard this over the weekend, but I don't have the LT half-bridge driver chip. It's not ideal anyway as it only allows a maximum of 60V on the topmost drain. The whole point of this is to get something like a "cool" 100 Watts directed by the single-ended stage. I do have some TPS2817 drivers handy ~ I'll have to think of a way to float one up at the topside somehow.
I think the ripple results from lots of different factors in this circuit. Probably the hysteresis in the comparator is the biggest factor. Increasing the value of the inductor slows the rate of change in current injected by the switches, so for the same switch rate the ripple magnitude goes down with a bigger inductor. Of course this also reduces the slew rate into the load - which is where I think the linear amp was being called upon to fill the gap in that thesis paper. At this stage I was trying to keep the simulation minimal for the sake of clarity.A differential amp before the comparator should help free-up some of the parameters we're working with here. I was a bit constrained by the 0-5V range of the comparator inputs, and the resistive divider was an all-too-convenient solution really.
I'd like to breadboard this over the weekend, but I don't have the LT half-bridge driver chip. It's not ideal anyway as it only allows a maximum of 60V on the topmost drain. The whole point of this is to get something like a "cool" 100 Watts directed by the single-ended stage. I do have some TPS2817 drivers handy ~ I'll have to think of a way to float one up at the topside somehow.
The parallel amp is really off-topic to the Switching current source so I have initiated a new topic in the Class D sub-forum to keep things in their proper place. For my update on the weekend's experiment see:
http://www.diyaudio.com/forums/showthread.php?s=&threadid=148462
http://www.diyaudio.com/forums/showthread.php?s=&threadid=148462
I'd be disapointed if this subject were to be closed ...
I don't have experience with Class A design, so maybe my question is off-topic:
how do you bias the audio signal at the input of the A stage to 4.x volts in practise ?
Just an observation from the simulation:
The current through the flyback diode shows huge current spikes, some 30...50 amps. Similar voltage spikes at the drain of the switching Mosfet. Seems to occur when the current through the diode turns off. On the other hand these spikes dissapear when I select a different diode, like one of the motorolas. Maybe a flaw in the model of that particular component ? What does your breadboard say ?
In theory when the Mosfet turns off, the current through the L just continues to flow - but now through the diode - and hence still through he upper Mosfet. When the switch comes on again, the diode goes into reverse - dumping the charge from it's junction. This will cause some spikes but not that much - if a fast diode is selected.
When I tried to reduce the standing current by means of reducing the voltage at the comparator, I could not go much below 2 amps w/o loosing the lower part of the sine altogether. I had to increase the current sense resistor to twice it's value for 1 amp and even further for below 1 amp. Opamp / comparator offset ?
I don't have experience with Class A design, so maybe my question is off-topic:
how do you bias the audio signal at the input of the A stage to 4.x volts in practise ?
Just an observation from the simulation:
The current through the flyback diode shows huge current spikes, some 30...50 amps. Similar voltage spikes at the drain of the switching Mosfet. Seems to occur when the current through the diode turns off. On the other hand these spikes dissapear when I select a different diode, like one of the motorolas. Maybe a flaw in the model of that particular component ? What does your breadboard say ?
In theory when the Mosfet turns off, the current through the L just continues to flow - but now through the diode - and hence still through he upper Mosfet. When the switch comes on again, the diode goes into reverse - dumping the charge from it's junction. This will cause some spikes but not that much - if a fast diode is selected.
When I tried to reduce the standing current by means of reducing the voltage at the comparator, I could not go much below 2 amps w/o loosing the lower part of the sine altogether. I had to increase the current sense resistor to twice it's value for 1 amp and even further for below 1 amp. Opamp / comparator offset ?
payloadde said:
I am very pleased with the results I've managed to get from this. Compared to the parallel A/D I am much more confident that this is a worthwhile thing to build. If I was going to build a pure single-ended Class A right now, I'd definitely incorporate the switching current sink.
The single ended Class A would normally be AC coupled at the O/P as well as the input allowing a non-critical mid-rail setting (using a resistive divider to bias the gate) For the simulation only, to avoid charge-up time for the caps, I've DC coupled everything.
I think it's pretty clear that the main challenge of this circuit is to return the unused energy gracefully. I have been using some 31DQ10 diodes on the breadboard - actually, the switch/driver/inductor/diode/resevoir cap are all on a piece of double-sided FR4 (cut into islands for the parts) in lieu of a proper PCB. The inductance of connections to the diode are critical. I'm not sure how this plays in the LTspice world! I am sure there are much better diodes for this job.
I set the gain of the error amp to make it clear (1v per amp) what was going on. This can be further optimised. The ultimate goal is to track the signal and apply an appropriate bias (by dynamically varying the comparator threshold voltage) so that the dissipation is further reduced. I am also planning to use an inexpensive CPU cooler block and temperature control the fan speed as well.
The ultimate goal is to track the signal and apply an appropriate bias (by dynamically varying the comparator threshold voltage) so that the dissipation is further reduced.
What I meant was, that the oscillator behaves peculiar when I try to reduce the standing current by reducing the reference voltage at the comparator. In my setup 2 amps seem to be just O.K. But somewhere between 2 and 1 amps, strange things happen. To analyse it a bit further, I replaced the audio driver FET with a 6 ohm resistor. Sometimes below 2 amps either the frequency suddenly jumped to 1/2 the nominal, or the signal at the gate of the FET switch showed a sequence of pulses with different length - like 1st pulse medium length, 2nd short, 3rd extra long a.s.o. I don't know why it does that, but it may not be so easy as to just reduce the reference voltage at the comparator to save power. To get rid of the anomaly at lower current I had to also adjust the sense resistor, i.e. increase it. This is of course not practical. Maybe there is just too much noise across the sense resistor ...
B.t.w. if you tune the standing current, you also have to tune the bias at the gate of the audio driver FET - otherwise you cannot keep the midlevel at the output.
To play it safe and since my pre-amp does not yield more than 15v(pp) and the source follower having unity gain anyway, I built my test circuit with just 30v split supply. The switch driver is just a complementary pair of BJTs - everything out of my junk box ...
What I meant was, that the oscillator behaves peculiar when I try to reduce the standing current by reducing the reference voltage at the comparator. In my setup 2 amps seem to be just O.K. But somewhere between 2 and 1 amps, strange things happen. To analyse it a bit further, I replaced the audio driver FET with a 6 ohm resistor. Sometimes below 2 amps either the frequency suddenly jumped to 1/2 the nominal, or the signal at the gate of the FET switch showed a sequence of pulses with different length - like 1st pulse medium length, 2nd short, 3rd extra long a.s.o. I don't know why it does that, but it may not be so easy as to just reduce the reference voltage at the comparator to save power. To get rid of the anomaly at lower current I had to also adjust the sense resistor, i.e. increase it. This is of course not practical. Maybe there is just too much noise across the sense resistor ...
B.t.w. if you tune the standing current, you also have to tune the bias at the gate of the audio driver FET - otherwise you cannot keep the midlevel at the output.
To play it safe and since my pre-amp does not yield more than 15v(pp) and the source follower having unity gain anyway, I built my test circuit with just 30v split supply. The switch driver is just a complementary pair of BJTs - everything out of my junk box ...
Attachments
Since I got similar irregular pulse trains with 2 amps and 10 ohm resistor, which is about the situation with a negative swing of the audio input signal, I decided to use a current mode PWM chip which runs at a pre-selected frequency, although qed047 said earlier that he didn't like it because PWM tends to increase ripple. I found a venerable SG3842 aka UC3842 aka LT1242 in my junk box and the data sheet says it can run at 500kHz, so I gave it a try. It has a peak current sense feature which requires an additional sense resistor, a built-in comparator and an error amplifier with a fixed 2.5v reference. The error amp feedback input gets the signal from the same sense amp as before. Total power consumption from the supply is 36W, 33W burnt in the audio driver, hence 3W for the CCS including all gear. 6W into 6 ohm is all there is, but fills a small room with music. Drawback is, adjusting the standing current depending on volume is not so easy because the comparator reference is fixed at 2.5v inside the chip. The only way I can think of is to adjust the gain of the sense amp. for that purpose. I also did some experiments with a crude DC servo, but not shure whether it is completely stable ... Actually didn't really need it, the midpoint level stayed tuned to within 200mv all the time.
Attachments
Silly me ... bad idea to manipulate the current feedback for the purpose of DC servo. It is of course much better to control the bias of the audio FET. This also solves the above mentioned problem of keeping the mid level when the standing current is changed depending on volume. To do that I still have no other idea than to manipulate the sense amp to accomplish that ...
Attachments
Good to see you experimenting with the switching current sink payloadde. A few additional thoughts - the four resistors around the sense amp (10k & 1K) need to be matched to within 1% or better to ensure proper behaviour... the CMRR of the amp here is critical as the ends of the sense resistor bounce up and down between the supply rails. Also the slew rate of the opamp really could do with being a bit higher. I couldn't find a good datasheet for the discrete comparator shown in post 86 ~ maybe it's response time is too long?
Too bad the PWM chip uses a fixed reference. I would persevere with the discrete approach myself... starting with the four resistors (I had a whole bunch of 1K and 10K and matched two pairs with a ohmmeter).
Too bad the PWM chip uses a fixed reference. I would persevere with the discrete approach myself... starting with the four resistors (I had a whole bunch of 1K and 10K and matched two pairs with a ohmmeter).
I understand what you mean, however there is a definite listening pleasure to be had from a single-ended output. Class A-o-philes will happily argue that point all day.
The price of heatsink is far higher than the cost of parts needed to replace it using the current switching circuit we have here and although it may look rather foreign in an amplifier, if you take everything from the sense resistor down and put it in a "black box" labelled "current sink" it does no violence to the purity of a single-ended o/p stage. In fact, this is literally how I would recommend configuring a practical implementation of the idea.
Lumba Ogir, you seem to be determined to discourage us. if you're talking from experience then it would only be fair to let us know what practical issues you are referring to. My own experience is that switch-mode can deliver highly worthwhile audio amplification and that it can be both fun and profitable to experiment with these readily accessible techniques.
if you're talking from experience then it would only be fair to let us know what practical issues you are referring to. My own experience is that switch-mode can deliver highly worthwhile audio amplification and that it can be both fun and profitable to experiment with these readily accessible techniques.
Well, speaking for myself, for me this is just playing around. Proof of concept kind of thing.
I actually found a solution to adjust the standing current with my current mode PWM regulator chip. For that purpose I increased the gain of the sense amp to 30x and added a variable voltage devider to it's output with a range 1/1 ... 1/3. This gives a standing current variation between 0.7A ... 2A. This potentiometer is mechanically coupled with the volume control. Needs a 4-way pot for stereo then. The DC servo perfectly keeps the midpoint voltage, no need for a cap at the output.
Again, just a demonstrator. For a serious project this would need to be scaled up a bit. 60 volts +, 5 amps +, SMPS power supply, output filter, pre-amp.
I actually found a solution to adjust the standing current with my current mode PWM regulator chip. For that purpose I increased the gain of the sense amp to 30x and added a variable voltage devider to it's output with a range 1/1 ... 1/3. This gives a standing current variation between 0.7A ... 2A. This potentiometer is mechanically coupled with the volume control. Needs a 4-way pot for stereo then. The DC servo perfectly keeps the midpoint voltage, no need for a cap at the output.
Again, just a demonstrator. For a serious project this would need to be scaled up a bit. 60 volts +, 5 amps +, SMPS power supply, output filter, pre-amp.
Attachments
QED047,
Not at all, on the contrary. You should pay greater attention to signal integrity. Every part will effect (harm) the audio signal. You should also think more in the time dimension. Every diode junction introduces a time delay, precluding accuracy.
It depends on the definition of worthwhile. If you mean efficient audio amplification, I can agree.
That's the spirit
I hope you're getting a big kick out of seeing a heatsink-less transistor doing the job down below! But what are your impressions about the sound you get? I've done some critical blind listening tests using QED junior's *very sensitive* hearing equipment and there were no significant correlations to be had!
What about the sound ?
Well, it seems that this little (s)witching devil fools me into believing that
a) I hear all the music there is
b) I cannot hear any of the artifacts from the rude treatment in the cellar.
After all we are chopping several amps of current 400,000 times a second into pieces and then of course I've got this terrible pain in all the diodes down my left-hand side ... ;-)
so where's the catch ? there must be ! THD could be terrible if we decided to measure it.
In one of your previous posts you wrote:
the four resistors around the sense amp (10k & 1K) need to be matched to within 1% or better to ensure proper behaviour...
May I ask what were the symptoms before matching them ?
Well, it seems that this little (s)witching devil fools me into believing that
a) I hear all the music there is
b) I cannot hear any of the artifacts from the rude treatment in the cellar.
After all we are chopping several amps of current 400,000 times a second into pieces and then of course I've got this terrible pain in all the diodes down my left-hand side ... ;-)
so where's the catch ? there must be ! THD could be terrible if we decided to measure it.
In one of your previous posts you wrote:
the four resistors around the sense amp (10k & 1K) need to be matched to within 1% or better to ensure proper behaviour...
May I ask what were the symptoms before matching them ?
Nice, you made it with safer Vgs area from ripple when compared to that original schematic at page 1.
Look at here, just for references. Another liear+switching mixture
http://www.diyaudio.com/forums/soli...opology-where-diy-projects-2.html#post2223627
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.