Balancing Act
The next schematic demonstrates the idea - I built it and guess what ? It actually WORKED!!
So the cunning
plan seemed to be falling in place
- the input LTP provides some amplification (half of what a single ended stage might since it was a LTP) and low-ish impedance to drive the output fets.
- the current through R1 and R4 multiplied by their resistance provides the necessary bias voltage
and we have done that with the separate or floating rails that the output stage needed. Plus ..
- by grounding the unused input, it would also do duty as a phase-splitter.
I played for a time with various combinations of standing current and load resistance to arrive at what seemed optimal but frankly i may as well not have bothered - the final values i ended up with were pretty much what nelson had all along - a low couple of hundred ohms as the load and as much current as is needed to get the bias voltage right.
The choice of fets however was another matter - but i told you that already earlier on in this thread. The IRF 9610s worked OK but I found a few other parts that worked better. Suffice it to say that the input and output stages had to be looked at as a system & altering either affected the final amp in many interesting ways. I say interesting because it demonstrated how you really had to design around the parts you want to use. For example, using low threshold voltage jfets/mosfets/latfets just weren't ideal if i wanted to keep the beefy To-220 input stage.
Which I did as they provide enough drive to a) ameliorate distortion from the varying Ciss of the output fets and importantly b) so i could parallel fets later without worrying about drive current ability. I guess another reason was that it was just easier - kind of a blunt sledgehammer approch. So another approach would perhaps be the more sensitive one - with jfets as the input LTP and perhaps the semisouth R100s as the outputs. Like a circlotron Aleph J. The fact that neither of these are available anymore (i have a stash of both devices but thought better than to design a ground up amp around them) sealed the deal.
Note that I am a slow worker and and I learn things by experimenting, writing copious notes and taking photos and then thinking about what was goig on. Then testing again to see if the understanding was correct. Not much of a surpise then that the initial design first phase upto this point took about six months.
In attempting to put the front end and o/p stage together, false notions of cleverness hit. I tried doing the more interesting part first - making it work well with unbalanced (single-ended) inputs.
And so it was that I wasted another three months chasing down that proverbial wild-goose. Posts #1-12 tell a bit of the story.
Next up - a clean slate, redemption (of sorts...) and more about those resistors we use to make the signal 'ground' and what happens if you get them wrong.
PS- yes a very nice amplifier comes out at the other end - eventually
The next schematic demonstrates the idea - I built it and guess what ? It actually WORKED!!
So the cunning
plan seemed to be falling in place - the input LTP provides some amplification (half of what a single ended stage might since it was a LTP) and low-ish impedance to drive the output fets.
- the current through R1 and R4 multiplied by their resistance provides the necessary bias voltage
and we have done that with the separate or floating rails that the output stage needed. Plus ..
- by grounding the unused input, it would also do duty as a phase-splitter.
I played for a time with various combinations of standing current and load resistance to arrive at what seemed optimal but frankly i may as well not have bothered - the final values i ended up with were pretty much what nelson had all along - a low couple of hundred ohms as the load and as much current as is needed to get the bias voltage right.
The choice of fets however was another matter - but i told you that already earlier on in this thread. The IRF 9610s worked OK but I found a few other parts that worked better. Suffice it to say that the input and output stages had to be looked at as a system & altering either affected the final amp in many interesting ways. I say interesting because it demonstrated how you really had to design around the parts you want to use. For example, using low threshold voltage jfets/mosfets/latfets just weren't ideal if i wanted to keep the beefy To-220 input stage.
Which I did as they provide enough drive to a) ameliorate distortion from the varying Ciss of the output fets and importantly b) so i could parallel fets later without worrying about drive current ability. I guess another reason was that it was just easier - kind of a blunt sledgehammer approch. So another approach would perhaps be the more sensitive one - with jfets as the input LTP and perhaps the semisouth R100s as the outputs. Like a circlotron Aleph J. The fact that neither of these are available anymore (i have a stash of both devices but thought better than to design a ground up amp around them) sealed the deal.
Note that I am a slow worker and and I learn things by experimenting, writing copious notes and taking photos and then thinking about what was goig on. Then testing again to see if the understanding was correct. Not much of a surpise then that the initial design first phase upto this point took about six months.
In attempting to put the front end and o/p stage together, false notions of cleverness hit. I tried doing the more interesting part first - making it work well with unbalanced (single-ended) inputs.
And so it was that I wasted another three months chasing down that proverbial wild-goose. Posts #1-12 tell a bit of the story.
Next up - a clean slate, redemption (of sorts...) and more about those resistors we use to make the signal 'ground' and what happens if you get them wrong.
PS- yes a very nice amplifier comes out at the other end - eventually
Attachments
Last edited:
Mork Calling Orson
So far, we have seen that both the front-end and ouput stages are really similar to circuits that are already familiar and popular. Before we put the whole thing together though, we should take a closer look at one thing which is different in this circuit - the 'virtual' earth connection.
The amplifier needs a common signal reference in order to work - the outputs for example need to be compared against the inputs via feedback and to do this we need a common point to enable the comparisons to be made.
We call this the signal earth and in most typical amps, there is a common 0 volt rail that can be conveniently used for this purpose. We of course cannot, since there is no common points in the supply for the circ. So the two resistors are used to create one.
Take a look at the usual common source output stage that we are familiar with, attached below. The source pins of the output mosfets are at the same AC potential as the supply zero volt rail and this is our signal ground (we assume that the voltage rails are dead shorts at AC)
If you put a resistance between signal ground and the zero volt point, you start to 'float' the source pins a bit and if the resistances are big enough, the result is noise and hum.
Now take look at the schematic in post #18. Now trace back the source pins of the fets back to ground (again assuming the voltage supplies are AC shorts). Looks like there's a 100R resistance now in each of these paths. A big deal ? Well yes and no. To make sure that we don't get much noise injected to our amp- we want to keep this resistance low. But if we make it very low, the dissipation in them starts to become material. If we make it too high, dissipation porblems go away but we will have invited unwelcome noise into our otherwise quiet amplifier.
It turns out that the value is not too critical in the 100s of ohms and as we'll see later, if we use slightly highish values, the balance of the two resistors can be varied via a pot in order to tweak even-harmonics. Notably it accomplishes this task without upsetting the amps operating points or offset.
The next post shows the effect of going from 100R to 1K each for the two resistors.
So far, we have seen that both the front-end and ouput stages are really similar to circuits that are already familiar and popular. Before we put the whole thing together though, we should take a closer look at one thing which is different in this circuit - the 'virtual' earth connection.
The amplifier needs a common signal reference in order to work - the outputs for example need to be compared against the inputs via feedback and to do this we need a common point to enable the comparisons to be made.
We call this the signal earth and in most typical amps, there is a common 0 volt rail that can be conveniently used for this purpose. We of course cannot, since there is no common points in the supply for the circ. So the two resistors are used to create one.
Take a look at the usual common source output stage that we are familiar with, attached below. The source pins of the output mosfets are at the same AC potential as the supply zero volt rail and this is our signal ground (we assume that the voltage rails are dead shorts at AC)
If you put a resistance between signal ground and the zero volt point, you start to 'float' the source pins a bit and if the resistances are big enough, the result is noise and hum.
Now take look at the schematic in post #18. Now trace back the source pins of the fets back to ground (again assuming the voltage supplies are AC shorts). Looks like there's a 100R resistance now in each of these paths. A big deal ? Well yes and no. To make sure that we don't get much noise injected to our amp- we want to keep this resistance low. But if we make it very low, the dissipation in them starts to become material. If we make it too high, dissipation porblems go away but we will have invited unwelcome noise into our otherwise quiet amplifier.
It turns out that the value is not too critical in the 100s of ohms and as we'll see later, if we use slightly highish values, the balance of the two resistors can be varied via a pot in order to tweak even-harmonics. Notably it accomplishes this task without upsetting the amps operating points or offset.
The next post shows the effect of going from 100R to 1K each for the two resistors.
Attachments
A quick note about the measurement system I use - its a Prism Audio Dscope III (yeah catchy name I know). See nawavguys blog NwAvGuy: Testing Methods for a roundup but basically its a (very slightly cheaper) alternative to the AP analyzers and comes with pretty nifty features for R&D type work. I like it a lot but still keep my AP Portable 1 DD and a Pana VP7723 for some measurements.
Anyway, the THD spectrum on the left is made using 100R resistors and the right shows the spectrum using 1k resistors.
Nothing too earth shattering here and as we speculated above - only the noise is affected. The distortion products are pretty much unaffected (H2 -102db, H3-105db) and THD is 0.001 in both cases. But the noise has gone up a bit with the higher resistors as shown by the increase in THD+N to 0.005%. In the second spectum, we can also see that noise and PSU harmonics are increased. Not by much but enough to notice on a scope. The power op was abt 2w into an 8R load for these measurements at an 80khz bw.
Ok so this tells us what we need to know about the earth resistor network, and in the next few posts I'll show the schematic, interesting resemblences and some measurements from the prototypes. btw to the 40 or so people who are following this - thanks for reading
Anyway, the THD spectrum on the left is made using 100R resistors and the right shows the spectrum using 1k resistors.
Nothing too earth shattering here and as we speculated above - only the noise is affected. The distortion products are pretty much unaffected (H2 -102db, H3-105db) and THD is 0.001 in both cases. But the noise has gone up a bit with the higher resistors as shown by the increase in THD+N to 0.005%. In the second spectum, we can also see that noise and PSU harmonics are increased. Not by much but enough to notice on a scope. The power op was abt 2w into an 8R load for these measurements at an 80khz bw.
Ok so this tells us what we need to know about the earth resistor network, and in the next few posts I'll show the schematic, interesting resemblences and some measurements from the prototypes. btw to the 40 or so people who are following this - thanks for reading
Attachments
Last edited:
way more than i'd like to admit publicly zm ...
For $200 bucks though, the quantasylum QA400 comes with great endorsements - see the thread over here .... http://www.diyaudio.com/forums/equipment-tools/231401-quantasylum-qa400.html
For $200 bucks though, the quantasylum QA400 comes with great endorsements - see the thread over here .... http://www.diyaudio.com/forums/equipment-tools/231401-quantasylum-qa400.html
Here is the schematic I am using. We have just stitched together the circuit segments from earlier on with the addition of caps in parallel with the source bypass resistors - they allow the amp to deliver substantially more class-A watts for a given dissipation (square law action).
If these upset you, consider that they are in parallel with a low ohm resistor that we'd have lived with anyway....
If the circuit looks little familiar - well that coz it would be to most readers as i've intentionally arranged it to invoke similarity to a popular build.
It looks quite a bit like a Aleph-X but we have rearranged the PSUs and deleted the magic aleph current sources (along with some other protection bits). Not so strange looking this circ after all eh ?
The feedback arrangement was just a consequence of the circuit and not intentional - so whether it qualifies as super-symmetry (and therefore carry an "X" ) or not I guess only NP can say.
The input fets were 2SJ313 ($5bucks a go at current prices) or IRF9z24 ($1.55). Outputs were IXYS IXTQ36N30P ( $3.75). Rest of the bits are as cheap as chips.
Some measurements & specs coming up.
--------------------------------------------------
Edit: Error in schematic. See first post in this thread for latest version.
.
If these upset you, consider that they are in parallel with a low ohm resistor that we'd have lived with anyway....
If the circuit looks little familiar - well that coz it would be to most readers as i've intentionally arranged it to invoke similarity to a popular build.
It looks quite a bit like a Aleph-X but we have rearranged the PSUs and deleted the magic aleph current sources (along with some other protection bits). Not so strange looking this circ after all eh ?
The feedback arrangement was just a consequence of the circuit and not intentional - so whether it qualifies as super-symmetry (and therefore carry an "X" ) or not I guess only NP can say.
The input fets were 2SJ313 ($5bucks a go at current prices) or IRF9z24 ($1.55). Outputs were IXYS IXTQ36N30P ( $3.75). Rest of the bits are as cheap as chips.
Some measurements & specs coming up.
--------------------------------------------------
Edit: Error in schematic. See first post in this thread for latest version.
.
Attachments
My Curves
The input impedance (bal i/p) is 1.9K
Freq Response pretty wide (flat to 80kHz)
Phase response also flat (-2.5 degrees at 80K)
Power output graph shows both THD and THD+N. This is with 1.5A bias so the caps are giving us more class A than we would otherwise expect.
[just a note on this - some newer AP analyzers apparently strip out the noise at low levels]
1w THD is 0.009% THD+N 0.0025%. These are without any trimming btw.
THD+N vs freq is quite smooth.
Edit: Gain is 10x
Noise level of the prototype with regulated supplies is 40uV (20khz bw)
These are quite impeccable measurements by comparison to amps with many many more parts/stages (but they also have less heatsinking I'm sure!) ......so for an amateur and with few parts I am pretty happy. Of course, this is about as good as it going to get because we haven't dealt with the real-world inconveniences of PSU noise, banging it into an enclosure with toroids etc.
But there's (well hopefully) enough of a headroom here so if i stay within a few x of these, that'd do nicely!
The input impedance (bal i/p) is 1.9K
Freq Response pretty wide (flat to 80kHz)
Phase response also flat (-2.5 degrees at 80K)
Power output graph shows both THD and THD+N. This is with 1.5A bias so the caps are giving us more class A than we would otherwise expect.
[just a note on this - some newer AP analyzers apparently strip out the noise at low levels]
1w THD is 0.009% THD+N 0.0025%. These are without any trimming btw.
THD+N vs freq is quite smooth.
Edit: Gain is 10x
Noise level of the prototype with regulated supplies is 40uV (20khz bw)
These are quite impeccable measurements by comparison to amps with many many more parts/stages (but they also have less heatsinking I'm sure!) ......so for an amateur and with few parts I am pretty happy. Of course, this is about as good as it going to get because we haven't dealt with the real-world inconveniences of PSU noise, banging it into an enclosure with toroids etc.
But there's (well hopefully) enough of a headroom here so if i stay within a few x of these, that'd do nicely!
Attachments
Last edited:
Square Waves
If you are an electrostat fan that you will LOVE these... no more excuses for ringing
10kHz into 8R load... and then same again but but with 3uF in parallel. I havent shown 1k - pretty boring looking.
It cannot hope to match the F5 at the 100khzs but there you are.
Haven't had to tweak compensation but then again there's not as much feedback (so no output inductor) as is used in many class AB amps. I have a few small pf caps handy in case they are needed when build it into a chassis but stability looks very good so they may be able to rest easy in the drawer
If you are an electrostat fan that you will LOVE these... no more excuses for ringing
10kHz into 8R load... and then same again but but with 3uF in parallel. I havent shown 1k - pretty boring looking.
It cannot hope to match the F5 at the 100khzs but there you are.
Haven't had to tweak compensation but then again there's not as much feedback (so no output inductor) as is used in many class AB amps. I have a few small pf caps handy in case they are needed when build it into a chassis but stability looks very good so they may be able to rest easy in the drawer
Attachments
Last edited:
6.3V panny FM caps - look fairly dainty but pass lots of current and i used 2 in parallel to make sure.
yeah plenty of options since we only use N-ch fets and there are literally thousands. On my list i have the cree and rohm sics and a few interesting parts from toshiba and ST micro. The IRF240s were actually pretty good but the distortion rose quite a lot at higher frequencies - sounded like hifi though.
I blew a super-low capacitance RF power fet (and *&$# expensive too it was) just today so i've promised myself no more f around until i get this into a case and supplies - and for that I gotta do a PCB which i'm absolutely crap at. Asked for some help via PM but understandably folks who are good at this sort of thing are busy with the lovely AB amps on the SS forum.
Spent three hours playing with eagle pcb on a flight but couldn't even manage to lay down a resistor
I blew a super-low capacitance RF power fet (and *&$# expensive too it was) just today so i've promised myself no more f around until i get this into a case and supplies - and for that I gotta do a PCB which i'm absolutely crap at. Asked for some help via PM but understandably folks who are good at this sort of thing are busy with the lovely AB amps on the SS forum.
Spent three hours playing with eagle pcb on a flight but couldn't even manage to lay down a resistor
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.