New Lineup IDEA - Power Follower/Output stage

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Sakis I would not say its a bad practice at all, it depends on where the part is used and how. I always use fast parraleled drivers instead of one big slow one like mje15032 series. Harmon know exactly what they doing. ;)

Those BDs are exceptional as drivers although limited by Vce and SOA, two of them can be used in parrallel and they peform better than those toshibas at their max Vce especially at frequencies higher than 1Khz and lower powers.

Lineup is simulating not measuring, reality could be very different. His spice models for the BDs could be incorrect as well as the other models too.

agree .... no problems with your approach also ... Point is that my POV is most of the time coming from repair so i have seen the burned pcb and the thermals around some HK models ...plus he difficulty to find proper parts and re-stabilze them back ...Thank god this happened to only a few models

Kind regards sakis
 
Lineup,

If you don't have the capacity to physically test the design then I'll cobble one together on perfboard to see what it does. I don't know if it will start out as a BIGBT version right away, likely a triple EF since I have suitable parts for that handy. I will unfortunately be occupied with other obligations for a few days, hang tight unless someone else beats be to it.
 
Good News jkuetemann!
sakis and I are planing for a PCB.
He will make and order 4 PCB.
2 for himself and 2 for me.
But I tell him you may be interested in a PCB. Only if you want to, of course.
sakis has all the parts at home because he is an amplifier repairman.

So things progress.
When we have PCB we can see more guys come here and want to join the party.

Regards, Lineup

PS. The circuit will be Triple EF.
Like in the latest schematic I posted up here.
 
Hi Guys

I have a bread board of the triple-EF with CSs running, but lo and behold my scope has died! So, I can measure DC voltages but not see wave forms for a little while.

DC offset is a little tweaky - seems very dependent on device parameters. This was a detail Leach mentioned regarding why he did not use CSs or even current mirrors in his design. Matching of devices would help here. And Leach had a point about turn-on thumps when CSs are used, and although his observations were based on late 1970s device performance, his observations are correct. Most people will tolerate turn-on/off thumps, but I do not. Potential damage to the speaker is only one issue; the sharp transient is simply unsettling to hear.

I never like designs that require an offset adjustment. What is the guarantee that the adjustment will be correct a year later? Ten years later? I prefer to make the design itself take care of the offset, and most methods require adding some more BJTs.

One approach is to use a second diamond buffer tied to the output but have its collectors tie into the current sources for the input stage. Depending on what compensation is used in the final circuit form, one may or may not need an RC to filter out the AC signal, between the speaker output and this buffer's input node.

Another approach uses complimentary diff-amps with one side grounded and the other with a filtered version of the speaker output. The appropriate collectors are tied in to the CSs for the input stage. This approach has a lot more gain to correct for larger offsets. However, the basic circuit does not seem to exhibit much offset except when mosfets are used. Still, that offset is higher than I like in an amp.

With the CSs overall response is wider and gain is higher, so layout becomes critical. The symmetry of the circuit suggests a very symmetrical and clean layout. The layout can become tricky with more pairs of outputs if you are going for widest bandwidth. For top performance, it is not just an issue to match devices, but also compensation caps. Cap tolerances are pretty wide, and even if you can get 1-2% discs, mica or polystyrene, you still need to select matched pairs for a given channel assembly.

Have fun
Kevin O'Connor
 
What do you think Struth,
about how to put transistors on the main heatsink.
Should the drivers be put there, too.
How did Mr Leach do with his triple follower.

As I told sakis is making PCB.
We need some advice.
I told him not to put the drivers onto main heatsink .. but I am not sure.
 
U6,7 and U3,9 should be thermally coupled to prevent large DC offset

For pre-drivers depends on their heat dissipation value if they can stay away from the main heatsink
Yes, I told sakis those 4 transistors should be close, if not in touch with eachother.

For the 2SC4793/SA1837 drivers, they have like 50-55 mA running.
So they do get a bit hot.
But I have decided to put them on smaller separate heatsinks.

Only the Output transistors MJL3821/1302A together with the BD139 vbe multiplier on the Main heatsink.

Do you think this will be good?

My reference is the wellknown M. Leach lowtim amplifier
which uses Triple EF figuration of output stage.
The Leach Amp - Output Stage

:)
 
Not only close, pair's flats faces tightly bonded together.

For triple output Leach approach is just fine.
Thanks Lazy Cat.
I tell him.


can we use DC-servo?
Of course one could add DC-Servo.
But I dont think we should.
There are other ways more simple.

If you look at my latest schematic in this post: http://www.diyaudio.com/forums/soli...ower-follower-output-stage-3.html#post3242277
you see one trimpotentiometer on the input.
This is the way to adjust for the output offset.

Also one could adjust the current sources of the transistors.
 
OK ....first attempt for pcb is done
here is some detailed pictures and also large files can be downloaded from my server .
There is also detailed explanation and also someone should double check for errors
Users that also have S-Print layout 4 may also down load a complete file from my server
board size is 135X80

kind regards
sakis

An externally hosted image should be here but it was not working when we last tested it.


An externally hosted image should be here but it was not working when we last tested it.


An externally hosted image should be here but it was not working when we last tested it.


An externally hosted image should be here but it was not working when we last tested it.
 
Hi Guys

The drivers should never be on the main heat sink; they should always have their own board-level heat sink.

As LC stated, the diamond input devices need to be in contact with each other to thermal track, not only from one half of the circuit to the other, but within each half. To make this simpler mechanically, all four BJTs in the diamond should be 2SC/2SA, then all four can be bolted together or at least bolted to a common piece of flat aluminum supported by the four BJTs.

Obviously there needs to be thermal tracking for the usual bias generator forms. Two-fold tracking is usually best, where ambient is tracked by one device and the output transistor temp by another device. For a CFP output, the driver would be tracked instead of the outputs. (For a CFP with gain, both need tracking).

Again with the offset trim: adjusting input node resistance seems tenuous as it makes the input impedance for the circuit halves different. This could effect performance, especially at high frequencies. Adjusting a current source on a one-time basis using a trimpot may work, but as with any trimpot approach the question still remains as to long-term reliability of the setting's effectiveness.

An op-amp based servo introduces its own issues. The discrete servo options outlined earlier are simple and have an appeal because they are discrete and can operate directly from the supply rails.

There should be local filter caps on the PCB. This improves transient response and reduces the interaction of the amp with the supply. A small bypass cap should go rail-to-rail to make the rails look "more identical" at very high frequencies. Bypasses across each filter cap are not always a good idea and can inject noise into the ground path.

Does your board supplier charge more for double-sided versus single-sided? The standard these days is double-sided, so there should be no need for jumpers. Also, get 2-ounce copper as a minimum. This used to be standard but now the default is 1-ounce unless you specify it.

Traces for the supply feeds to the outputs, the connections to the speaker, ground paths between the filters and speaker, should all be fairly wide. How wide depdns on the current to be handled. There are charts available on the net that show treack width versus copper weight and temperature rise. For example, 100Wrms output at 8R represents 3A5 rms or 5Apk. To be conservative, take the peak value. For a temperature rise of the trace limited to 10C, and 2oz copper, the track width should be about 60mil - 0.060" (1mil=0.001" - a North American machining term).- or about 1.5mm.

The tracck width example assumes a purely resistive load and a higher actual current may be present with a real speaker load, so make the high-current tracks wider if possible.

The front-end track width should be narrow for best high-frequency performance and not have any right angles. Even orthogonal traces bent at 45-degrees can be problematic as the sharpness of the corners can act like antenna. Some PCB software has a feature to round these, otherwise it must be done manually. The compromise with skinny traces is that they may not take much resoldering as usually the pads around the holes are also skinny.

Supply connections (+,0,-) should accommodate #18 wire. So should the speaker connections (out,0). The input should accommodate shielded wire, which might require a large-ish hole for the shield connection depending on the type of coax you wish to allow for. Input and output grounds should be decoupled.

Have fun
Kevin O'Connor
 
I will agree with most of the remarks with Kevin ...

Still Kevin this is just a demo board to make it run ...so from all the above the only thing that is really important for the operation of the amplifier is the thermal matching issue .

Line up has to give me directions about that and about which devices should be coupled with which ...

As about a diamond with all TO220 devices i think is a bad approach ...

Kind regards
sakis
 
I will agree with most of the remarks with Kevin ...

Still Kevin this is just a demo board to make it run ...so from all the above the only thing that is really important for the operation of the amplifier is the thermal matching issue .

Line up has to give me directions about that and about which devices should be coupled with which ...

As about a diamond with all TO220 devices i think is a bad approach ...

Kind regards
sakis
AS Lazy Cat said here,
those 4 transistors in the diamond should be together 2 and 2.
First the two 2SC4793+2SA1837
and also the two 2N5551+2N5401.
It is also good if those two pairs are very close to eachother.

Another thing that was wrong in the schematic I emailed you, sakis
is the output filter.
Inductor+resistor comes first and at the end comes capacitor+resistor(Zobel).

You can see it in this picture:
outstage1.gif
 
Hi Guys

I've used the tabbed BJT solution for thermal tracking and it works fine. It is a mechanical convenience. The IC forms of the diamond buffer, like the LT1010 and similar, rely on the thermal tracking of all four devices to provide both wide response and any semblance of acceptable DC performance. It helps that the devices are fabricated on the same die and hopefully have similar specs.

Without close tracking of gains with temperature, you have to take a hit on bandwidth and increase degeneration. This allows less accuracy of all types of matching to be less problematic. For a chip that is supposed to be a video driver, such a solution is unacceptable. For a discrete circuit that is to handle audio frequencies, it is purely an aesthetic affront to have performance diminished from the simulated theoretical ideal.

This buffer is a wideband circuit, so the details in my post are not just refinements for a later date, but may adversely effect initial results without them being in place from the beginning. Despite the compensation caps, oscillation due to layout error is a possibility with this circuit. If you are going to the trouble to etch your own boards, or to order them from someone else, why not lay out as much of the final solution as possible? These are aspects of board layout that apply to every audio PA design, not just to this one. Lineup asked for opinions and direction about thermal tracking and basic layout. You can take or leave any advice offered.

I take a conservative approach when testing and designing. I haven't blown up devices during an initial test since I was in high-school and knew much less than the little I know today.

Have fun
Kevin O'Connor
 
Last edited:
AS Lazy Cat said here,
those 4 transistors in the diamond should be together 2 and 2.
First the two 2SC4793+2SA1837
and also the two 2N5551+2N5401.
It is also good if those two pairs are very close to eachother.

Another thing that was wrong in the schematic I emailed you, sakis
is the output filter.
Inductor+resistor comes first and at the end comes capacitor+resistor(Zobel).

You can see it in this picture:
outstage1.gif

Most of the power amp zobels use RC connected close to the output and inductance part could be close to the loudspeaker terminal. D. Self does it that way and many others.
dado
 

Attachments

  • zobel.jpg
    zobel.jpg
    159.7 KB · Views: 191
Again with the offset trim: adjusting input node resistance seems tenuous as it makes the input impedance for the circuit halves different. This could effect performance, especially at high frequencies. Adjusting a current source on a one-time basis using a trimpot may work, but as with any trimpot approach the question still remains as to long-term reliability of the setting's effectiveness.

As I noted, I also think that it will change the gain on both transistors. Is there any one going to react to this?
 
Hi Guys

Convention is not necessarily the best way. Leach had output stages oscillate to death with the zobel on the amp side of the coil. If you are tired of Leach, look at Bryston: they place the zobel after the choke, too. Chris Russell and Stuart Taylor are both very smart and tuned-in designers, as is Marshall Leach.

I think even Archie Leach would agree.

The trimpot for DC offset definitely will be problematic in its present position. The concern about performance is real. A less crucial facet is the fact both signal and DC currents are effected by the pot's rotation, so definitely adjust it WITHOUT a speaker in place - unless you like earthquake noises.

Considering the effort so far to create a power buffer with stupendous performance - wide bandwidth, super-low THD - it seems like a major fumble to retain the input offset trimpot. FdW and I seem in agreement on this, but Sakis will only consider Lineup's input for the PCB. His position is understandable considering his effort to design the card, but he has already acknowledged that this is only a beginning: more board designs will be necessary.

Have fun
Kevin O'Connor
 
Last edited:
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.