The previous Posts bandwidth can be increased, by 4 times to 16 MHz by not using a CFP output stage. Instead, the LT1223 add on stage is boosted for 50mA output by paralleling five 2N5550's and 2N5401's and the output stage is as a unity gain voltage follower.
The open loop gain of the LT1223 is determined by the inverting input resistor R18 which is reduced to 220 Ohms (was 2k2). The THD at 1W is now 0.6ppm (was 0.3ppm).
Also, with this modification the THD at 1MHz is much reduced -- without the CFP's sluggish turn-off limitation.
A better topology choice for a fast nonswitching power amp👍. And about the same number of 2N5550's and 2N5401's and their heat dissipation.
Again, when the feedback is taken from "Op1" instead of "Out" there is no global feedback so the amp is less sensitive to reactive loads (like here) but THD is significantly higher (but still probably unnoticed with audio):
. . . . 1kHz 1W . . 20kHz . . . BW
GFB . 0.6ppm . . 1.3ppm . . 16MHz
NGFB . 50ppm . . 46ppm . . 7MHz
The open loop gain of the LT1223 is determined by the inverting input resistor R18 which is reduced to 220 Ohms (was 2k2). The THD at 1W is now 0.6ppm (was 0.3ppm).
Also, with this modification the THD at 1MHz is much reduced -- without the CFP's sluggish turn-off limitation.
A better topology choice for a fast nonswitching power amp👍. And about the same number of 2N5550's and 2N5401's and their heat dissipation.
Again, when the feedback is taken from "Op1" instead of "Out" there is no global feedback so the amp is less sensitive to reactive loads (like here) but THD is significantly higher (but still probably unnoticed with audio):
. . . . 1kHz 1W . . 20kHz . . . BW
GFB . 0.6ppm . . 1.3ppm . . 16MHz
NGFB . 50ppm . . 46ppm . . 7MHz
| A feature of this form of autobias spreader is BJT's can be swapped for MOSFETs like the IRFP240 and the bias setting is almost the same and requires only a slight trim for minimum distortion at say 1W. For example, the BJT version, trimmed for 300mA for minimum distortion at 1W, is replaced with IRFP240/IRFP9240's and the idle current is 250mA. To get the same minimum distortion at 300mA with these MOSFETs at 1W, the diode series resistances are increased slightly (from 75mR to 100mR). This is because MOSFET's have a slightly different transconductance curve than BJTs. Here MOSFET's are "paralleled" -- each with their own bias spreader and drivers while using a common opamp. Note the resistor on the opamp output needs to be reduced to provide sufficient drive current at the peaks since less booster transistors are used. The gate pull-up resistors can be increased without loss of bandwidth compared to the previous BJT version. Then booster current can be reduced allowing two (instead of five) driver pairs per output stage. This MOSFET version requires less parts while distortion at 1W is about the same as the BJT version and possibly higher BW (bench testing needed). BTW The parallel BJT version sim file is attached. And remember, the advantage with versions with opamp inputs is the ability to change to current drive😉. Pawel and Minek123 thanks for the likes. |
Hi All,
The Post 1 summary page has been updated to include Topology 3 (a CFP autobias stage) and Topology 4 (LV LT1223 opamp with the TO-92 voltage booster).
I have now designed 3 PCBs -- for Topologies 2 ,3 and 4. EG Topology 4 shown below:
BTW the power transistors fold down. Two pair can drive 200W into 4 Ohm loads. The PCB is 74mm H and 68mm W to fit Conrad heatsinks (Australian made).
The SOT-223 bias spreaders are visible. On the front there are SMD power diodes under the DO-201 TH diodes (not visible).
The bottom left has a 2W current-sene resistor for the speaker current drive option (it can be a wire link for voltage output).
The Post 1 summary page has been updated to include Topology 3 (a CFP autobias stage) and Topology 4 (LV LT1223 opamp with the TO-92 voltage booster).
I have now designed 3 PCBs -- for Topologies 2 ,3 and 4. EG Topology 4 shown below:
BTW the power transistors fold down. Two pair can drive 200W into 4 Ohm loads. The PCB is 74mm H and 68mm W to fit Conrad heatsinks (Australian made).
The SOT-223 bias spreaders are visible. On the front there are SMD power diodes under the DO-201 TH diodes (not visible).
The bottom left has a 2W current-sene resistor for the speaker current drive option (it can be a wire link for voltage output).
Dear Ian!
I am surprised that temperature sensing diodes are not connected directly to the HS... are they coupled with a copper trace soldered to output transistor?
kind regards
Pawel
I am surprised that temperature sensing diodes are not connected directly to the HS... are they coupled with a copper trace soldered to output transistor?
kind regards
Pawel
Hi Pawel,
Thanks. An important question others may be asking.
The autobias loop senses current in the power transistors using the power diodes. So there is no need to sense the temperature in the power transistors, just in the power diodes.
The bias spreader transistors, the SOT-223's seen on the back side (my previous post), need to sense the temperature of the power diodes.
The bias spreader transistors on the back side are directly under the power diodes on the top side. This gives rapid thermal coupling.
There is an SMD diode on the top side (red pads "Net D3 K"), and then the spreader Q15 (SOT-223 blue pads) is on the back side.
You can see a second TH diode in my previous post top side. Its in parallel with the SMD diode since each diode is rated at 3A and they need to handle up to 5A (peak).
Cheers, IanH
Thanks. An important question others may be asking.
The autobias loop senses current in the power transistors using the power diodes. So there is no need to sense the temperature in the power transistors, just in the power diodes.
The bias spreader transistors, the SOT-223's seen on the back side (my previous post), need to sense the temperature of the power diodes.
The bias spreader transistors on the back side are directly under the power diodes on the top side. This gives rapid thermal coupling.
There is an SMD diode on the top side (red pads "Net D3 K"), and then the spreader Q15 (SOT-223 blue pads) is on the back side.
You can see a second TH diode in my previous post top side. Its in parallel with the SMD diode since each diode is rated at 3A and they need to handle up to 5A (peak).
Cheers, IanH
wow, respect, extraordinary design!
next question: 2N5551s are TO92 cased, they dissipate >200mW, don't you think they will roast even if they are rated 650mW?
P
next question: 2N5551s are TO92 cased, they dissipate >200mW, don't you think they will roast even if they are rated 650mW?
P
Hi Pawel,
The 2N5551's are still available from digikey or Farnell. Hopefully for a few more years.
The dissipation was queried - see post 293-298
https://www.diyaudio.com/community/...ching-auto-bias-power-amp.375141/post-7880572
These TO-92's can handle 400mW with ambient air at 70C, but for safe operation say 200-300mW each. I'm not planning to add a small heatsink. Instead I'll parallel more since they are only 20 cents in 10's and I have enough board space. I hope to assemble a board to check temperatures.
The 2N5551's are still available from digikey or Farnell. Hopefully for a few more years.
The dissipation was queried - see post 293-298
https://www.diyaudio.com/community/...ching-auto-bias-power-amp.375141/post-7880572
These TO-92's can handle 400mW with ambient air at 70C, but for safe operation say 200-300mW each. I'm not planning to add a small heatsink. Instead I'll parallel more since they are only 20 cents in 10's and I have enough board space. I hope to assemble a board to check temperatures.
Hi Pawel,
If SMD diodes and SMD bias spreaders are intimidating to hand solder then it is possible to drill holes in the PCB to let TO-92 leads thru to contact TH diodes on the front side. Not the best for thermal linkage, but you could get my design to work with no SMD's.
I am assuming most DIY builders can hand solder SOT-223's and the SMD diodes. But not for beginners, you need to be good at soldering for that.
It's the 0805 and smaller, and even SO-8 opamps, that are best left to a heat gun or SMD oven.
If SMD diodes and SMD bias spreaders are intimidating to hand solder then it is possible to drill holes in the PCB to let TO-92 leads thru to contact TH diodes on the front side. Not the best for thermal linkage, but you could get my design to work with no SMD's.
I am assuming most DIY builders can hand solder SOT-223's and the SMD diodes. But not for beginners, you need to be good at soldering for that.
It's the 0805 and smaller, and even SO-8 opamps, that are best left to a heat gun or SMD oven.
Hi all!
I'm kinda new around here, and only just discovered this thread, which is weird, cos I'm sure I did a search for non switching when I joined the site. Ah well. I've only skimmed thru so far, there's rather a lot to take in but I can see there's many great ideas.
Anyway, I've independently started on my own NS path - the results of my efforts are documented here:
I'd certainly be interested in comments, as long as they're nice ones haha.
I'm kinda new around here, and only just discovered this thread, which is weird, cos I'm sure I did a search for non switching when I joined the site. Ah well. I've only skimmed thru so far, there's rather a lot to take in but I can see there's many great ideas.
Anyway, I've independently started on my own NS path - the results of my efforts are documented here:
I'd certainly be interested in comments, as long as they're nice ones haha.
Hi Minek,
If you look at the internal structure of the LT1223 CFA in its datasheet, the inverting input is an emitter node, so the inverting input resistor effectively determines the gm of the input stage before a feedback resistor is added. Therefore the amps overall open loop gain is set by the inverting input resistor to ground.
You can view this with an open loop ac plot. My circuit included an inductor in series with the feedback resistor for this. Note the shunt capacitor is still included for my semi "open loop" plot. Remove the capacitor to see the real native open loop plot.
If you look at the internal structure of the LT1223 CFA in its datasheet, the inverting input is an emitter node, so the inverting input resistor effectively determines the gm of the input stage before a feedback resistor is added. Therefore the amps overall open loop gain is set by the inverting input resistor to ground.
You can view this with an open loop ac plot. My circuit included an inductor in series with the feedback resistor for this. Note the shunt capacitor is still included for my semi "open loop" plot. Remove the capacitor to see the real native open loop plot.
Hi Binjo,
Nice for you to share your thoughts and designs. I did read your post.
If you read my first post it aims at a simple way to do an auto bias loop that is reliable and well behaved. Like most designers we want the auto bias to be non-switching - meaning in Class-AB one transistors current never drops to zero but remains slightly conducting (like 10mA or so).
Others have commented on this forum that non-switching is not easy to do, and especially as a simple circuit that sounds as good as the best currently available. With auto bias loops it is usually difficult to get them to be well behaved during clipping and/or running at the top end of their frequency range.
From what I read in your post you have been looking closely at all these issues and have found a way through many (maybe all) the difficulties. I wish you the best with making a prototype so you can hopefully iron out all the issues to demonstrate really good non-switching amp. We need more examples of working non-switching amps that also sound good👍,
Cheers, IanH
Ian, are you saying that you lowered R18 (and R19 along with it) from 2k2 (in previous version) to 220 Ohms, to lower open loop gain of the op-amp?
When I tried this in my amp, phase margin suffered - which would be weird if the OL gain decreased?
Hi Minek,
Lowering R18 increases loop gain. R18 2k2 gives 70dB below 1kHz and 96dB with 220 Ohms - 16dB difference. File attached.
The cursors are at 10MHz to measure phase difference. This is with the 4p7 overall feedback capacitor (normally across the feedback resistor).
Below is open loop without the capacitor:
Rolloff starts at 10kHz without this capacitor (was 1kHz). This capacitor is needed in closed loop with R18 of 220 Ohms to tame peaking above 10MHz. It gives about 16dB extra loop gain at 1kHz for a lower THD reading. But at 20kHz with C7 added the loop gain is about the same😀.
Also notice the effect of the capacitor (C7) on the output of the LT1223 in my circuit feeds the "Op1" node, where your posted circuits this capacitor goes to common. If you run my circuit (attached) closed loop you will see connecting C7 to ground instead of Op1 can give a wider (small signal) bandwidth. That's in simulation, but in reality parasitics will probably demolish these grandiose figures and compensation tricks.
I look forward to your findings if you decide to make a prototype using the LT1223. Will you use the DIP TH version?
Lowering R18 increases loop gain. R18 2k2 gives 70dB below 1kHz and 96dB with 220 Ohms - 16dB difference. File attached.
The cursors are at 10MHz to measure phase difference. This is with the 4p7 overall feedback capacitor (normally across the feedback resistor).
Below is open loop without the capacitor:
Rolloff starts at 10kHz without this capacitor (was 1kHz). This capacitor is needed in closed loop with R18 of 220 Ohms to tame peaking above 10MHz. It gives about 16dB extra loop gain at 1kHz for a lower THD reading. But at 20kHz with C7 added the loop gain is about the same😀.
Also notice the effect of the capacitor (C7) on the output of the LT1223 in my circuit feeds the "Op1" node, where your posted circuits this capacitor goes to common. If you run my circuit (attached) closed loop you will see connecting C7 to ground instead of Op1 can give a wider (small signal) bandwidth. That's in simulation, but in reality parasitics will probably demolish these grandiose figures and compensation tricks.
I look forward to your findings if you decide to make a prototype using the LT1223. Will you use the DIP TH version?
Thanks Ian.
Yeah, DIP 8 version. I'm waiting for the PCB to arrive. Can't do breadboard prototyping, too messy for me, and I always make mistakes.
But it's kind of expensive and time consuming to use PCBs for all prototypes - 50% of them don't work as expected and they go to trash 🙂
And I prefer TH devices - for esthetical, old school look, reasons 🙂
My C7 goes to the ground, your C7 goes to VAS (like Miller cap). I got better results and more stability with C7 going to ground...
Also, the resistor from op-amp output to the ground (R9 in my amp) - I was trying different values to maximize the bandwidth and stability.
It's optimal value depends on the what kind of drivers and output transistors I was using. Different for hexfets, and different for latfets.
As for my question about R18 - its value goes together with R19 and they BOTH determine overall gain.
Originally you had R19 at 22k, and R18 at 2k2. Later you lowered BOTH of them by a factor of 10, preserving the same overall gain of the amp.
That's what triggered my question. I thought it was maybe done to minimize the noise...
Yeah, DIP 8 version. I'm waiting for the PCB to arrive. Can't do breadboard prototyping, too messy for me, and I always make mistakes.
But it's kind of expensive and time consuming to use PCBs for all prototypes - 50% of them don't work as expected and they go to trash 🙂
And I prefer TH devices - for esthetical, old school look, reasons 🙂
My C7 goes to the ground, your C7 goes to VAS (like Miller cap). I got better results and more stability with C7 going to ground...
Also, the resistor from op-amp output to the ground (R9 in my amp) - I was trying different values to maximize the bandwidth and stability.
It's optimal value depends on the what kind of drivers and output transistors I was using. Different for hexfets, and different for latfets.
As for my question about R18 - its value goes together with R19 and they BOTH determine overall gain.
Originally you had R19 at 22k, and R18 at 2k2. Later you lowered BOTH of them by a factor of 10, preserving the same overall gain of the amp.
That's what triggered my question. I thought it was maybe done to minimize the noise...
Last edited:
I was PM'd by Pawel asking whether the bootstrapped auxiliary supply like in Post 302 output stage can be omitted by instead using a high-value pull-up resistors to the rails as below:
It works OK with MOSFETs and operates up to 70kHz with 3k pull-ups. But there is about 4V extra loss in the output swing -- it gives about 150W into 8 Ohms where as the bootstrapped auxiliary supply gives 200W into 8 Ohms. Splitting the pull-up resistors and adding bootstrap capacitors can give the full swing and 200W output.
Below is the output stage with bootstrap with 20V rails suitable for driving 4 Ohms with a single pair of power transistors
With four 1k2 resistors it is OK to 150kHz and delivers about 25W into 4 Ohms at 0.04% THD at 1kHz and 0.3% at 100kHz. At 1W 1kHz THD is 0.0004%.
The diode series resistors have been lowered to 50m Ohms to give a wider low-distortion region of 1.4W. Dissipation in each power transistor is 10W which is easily to heatsink.
Distortion is low enough without global feedback, so it can be fed from a standard opamp with 15V peaks to do a quick bench/listening test without a PCB.
Don't forget the spreader transistors need thermal linking to the diodes. For bench tests TH 2N2222/2907's are easier to use than SMD PXT222/2907's, and TH UF5401's than SMD ESBB's. And each diode is actually 2 diodes in parallel (NB m=2)😏 since the diode's currents peak at 5A.
Cheers, IanH
Below is the output stage with bootstrap with 20V rails suitable for driving 4 Ohms with a single pair of power transistors
The diode series resistors have been lowered to 50m Ohms to give a wider low-distortion region of 1.4W. Dissipation in each power transistor is 10W which is easily to heatsink.
Distortion is low enough without global feedback, so it can be fed from a standard opamp with 15V peaks to do a quick bench/listening test without a PCB.
Don't forget the spreader transistors need thermal linking to the diodes. For bench tests TH 2N2222/2907's are easier to use than SMD PXT222/2907's, and TH UF5401's than SMD ESBB's. And each diode is actually 2 diodes in parallel (NB m=2)😏 since the diode's currents peak at 5A.
Cheers, IanH
Hi All,
A PCB has been received and five are now at the SMD assembler:
This PCB allows Topology 1 (CE voltage gain with 6dB global feedback Post 254), and Topology 2 (discrete VAS Post 267), selectable with a jumper.
There are parallel stages allowing 4R load but intended for nominal 100W 8R systems. The PCB fits snugly to a Conrad 65mm heatsink (as shown).
There are several through hole resistors allowing trimming of idle current and emitter resistors and peak current to suit the power transistors HFE's.
A PCB has been received and five are now at the SMD assembler:
This PCB allows Topology 1 (CE voltage gain with 6dB global feedback Post 254), and Topology 2 (discrete VAS Post 267), selectable with a jumper.
There are parallel stages allowing 4R load but intended for nominal 100W 8R systems. The PCB fits snugly to a Conrad 65mm heatsink (as shown).
There are several through hole resistors allowing trimming of idle current and emitter resistors and peak current to suit the power transistors HFE's.