The Blomley Class B amplifier

[haiqu]

Some people are sensitive to hearing phenomena than others. Once these are pointed out to others these can be detected. Sibilants were a case in point with me and I learned that about 30 years when a friend listened to a couple of amplifiers that I had built auditioned on his speakers at his home. He mimicked a sibilant sound by mouth and I have hated overemphasized sibilants whether due to amplifiers, loudspeakers or source material ever since. At the bass end of the audio spectrum this has to keep time with the music and drums should start with a bap rather than boom. This is probably more due to speaker design as amplifiers should have good damping factors if the power supply is engineered properly. I have built capacitor coupled output amplifiers and direct coupled ones. My preference is for direct coupling. The only capacitor coupled amplifier I built was the JLH 1969 Class A amplifier which is still popular - in my view the 1996 direct coupled output update is a better job.

Yes. Someone once pointed out the quacking sound that polyester caps make to me and I've hated them ever since. And I agree that a good servo-controlled amp beats capacitor coupling hands down every day of the week. Case in point is the venerable Crown 300.

 

[dreamth]

Quacking sound of poliester caps? Never heard this expression before...Worth taking notes?!
You probably never heard a qsc amp ...By the way quad405 is a virtual ground class B design too!

 

[mjona]

Quacking sound of poliester caps? Never heard this expression before...Worth taking notes?!
You probably never heard a qsc amp ...By the way quad405 is a virtual ground class B design too!

Hmm. Could one say polyester capacitors "what the duck"? but not appropriate in any sense.

 

[Alexandre]

One of the main points of the circuit according to Linsley-Hood was to avoid temperature issues with the output stage. The problem of setting bias currents for the output stage is then removed to the splitters. I used BC546B and BC556B in simulations. These give acceptable but not stellar results at 1kHz with low output quiescent current. Unfortunately things fall apart at 20kHz.

Built or simulated?
I have only simulated, and I saw no reason to dismiss this very interesting amp. I think the performance indicated in Blomley's article is attainable. There are fundamental differences from a conventional amp, I'd point this out: distortion spectrum is fixed regardless of level. Whether Vout is 1V, 3V, or 10V the harmonics stay at the same relative level to the fundamental. For 10Khz the fifth harmonic is better than -70dB according to the article, and I have been able to verify this in the sim. You cannot exceed 900mV between bases of the splitters if you want to stay true to the article.

 

[haiqu]

I haven't heard them yet, and if it can beat the no-NFB servo JFET design I did 20 years ago I'll be mightily impressed. But I did buy 10 PCBs in expectation of good things, and I have lots of cousins and nephews to give them to if they're any good.

 

[Ruwe]

Built or simulated?
I have only simulated, and I saw no reason to dismiss this very interesting amp. I think the performance indicated in Blomley's article is attainable.

I can confirm that for this schematic simulation and reality correlate well. And just as you said the distortion doesn't vary with level. My scope's FFT is not really spectrum analyzer quality, but is good enough to show trends... The other good thing is that the distortion has a nice gradual fade-out of higher harmonics. 2nd and 3rd are the main visible components... I've seen much more complex spectra with other amplifiers.

 

[mjona]

Built or simulated?
I have only simulated, and I saw no reason to dismiss this very interesting amp. I think the performance indicated in Blomley's article is attainable. There are fundamental differences from a conventional amp, I'd point this out: distortion spectrum is fixed regardless of level. Whether Vout is 1V, 3V, or 10V the harmonics stay at the same relative level to the fundamental. For 10Khz the fifth harmonic is better than -70dB according to the article, and I have been able to verify this in the sim. You cannot exceed 900mV between bases of the splitters if you want to stay true to the article.

Simulated - like a lot of other amplifier circuits of good repute I have investigated. These evidently work as built and have graphs to support claimed performance, but fall short in THD results in simulations. I got started in Audio Electronics in 1976 when I attended evening hobby classes at the local secondary school and had found a Wireless World publication titled High Fidelity designs. For a good many of these semiconductor kits, printed circuit boards, and complete kits were sold by Powertran Electronics.

There were three packs one could buy for the Blomley design, pcb, resistors, capacitors and pots, and Semiconductor set. There were difficulties with making purchases from overseas suppliers. Things were more apt to go missing in the post so I passed on building it. As has been pointed out the original design was intended for use with 15 Ohm speakers and the circuit has high closed loop gain.
If this is reduced for digital sources which have higher output signals the closed loop gain could be reduced which would give an increase in the level of negative feedback.

As far as the splitter transistors go if these are not biased properly and each of these operate somewhat in the turn on region a lot would depend on switching speed of the devices used. While the output transistors don't turn off completely that does not get around the problem of stored base charges needing to be removed to get them to that low conduction level on opposite current cycles. This is a difficult balancing act for simulation purposes - if the amplifier is to work it would be easier to build it and move trim pots for that purpose.

 

[Alexandre]

As far as the splitter transistors go if these are not biased properly and each of these operate somewhat in the turn on region a lot would depend on switching speed of the devices used. While the output transistors don't turn off completely that does not get around the problem of stored base charges needing to be removed to get them to that low conduction level on opposite current cycles. This is a difficult balancing act for simulation purposes - if the amplifier is to work it would be easier to build it and move trim pots for that purpose.

Agreed, on this point (importance of the switching behaviour of the splitters). I think it's a good idea to put some pads on the pcb, similar to smd pads but larger, to try a few different parts in this position. And I would not trust what datasheets say, I need to verify.

About the outputs, I would argue that the "resting" side is very much on, even at peak output. You can adjust the amp so that they never go below 50mA, and this is without having to set bias much above that. How much benefit there is to the non-switching is open to discussion of course. But I find it very interesting.

 

[mjona]

Agreed, on this point (importance of the switching behaviour of the splitters). I think it's a good idea to put some pads on the pcb, similar to smd pads but larger, to try a few different parts in this position. And I would not trust what datasheets say, I need to verify.

About the outputs, I would argue that the "resting" side is very much on, even at peak output. You can adjust the amp so that they never go below 50mA, and this is without having to set bias much above that. How much benefit there is to the non-switching is open to discussion of course. But I find it very interesting.

Some research into stored charge effects in bipolar output stages will show what I am talking about. On a graph the trans-conductance slope Amps out per volts or gm in of a triple output stage is very steep which allows quite small base voltages to deliver very large currents. But the converse is true when it comes to reducing the voltage level to reduce the output stage current to 50 m.a. levels because the charges in the collector region massively outweigh those in the base. If you look at a series of hills to get over as if these were sine waves in a converse sense then it is less trouble to move a load downhill than it is to move it up again.

 

[Alexandre]

I'll have to read more about that, thank you.
IIRC this youtube channel had a video comparing switching speed of a few small signal bjts. It's not the video I'm attaching (which is also very relevant).

 

[Alexandre]

This might be an example of a good bjt for the splitter. 'ft' is 400MHz @ 500uA, and a whopping 2GHz @ 10ma. Switching time is specified and graphed. Only the beta is a bit low. Not sure if it has a complement, and it's probably hard to find. I'm just posting the data as a reference.

OTOH, the fellows that built the amp had excellent results with more common types, so the answer is to be found experimentally.

 

[Alexandre]

I didn't look at the voltage specs for the bjt, it only takes 15V. So this one is not suitable.

 

[uptick]

...maybe square law devices are better?

 

[dreamth]

I didn't look at the voltage specs for the bjt, it only takes 15V. So this one is not suitable.

And probably completely useless in audip equipment...

 

[Alexandre]

Those devices are probably not low-noise (one can never have it all in a bjt...). BUT, in that position, the noise doesn't matter, because it is a cascode (which is not operating in class A - it will switch if you follow the design by Blomley.) The beta is important in a cascode, and less than 100 is low. Finally, I think we've agreed that switching speed is very important for the "splitter", thus the device suggestion (which didn't work out because of the voltage rating).

You could alter the circuit so there's no switching at all, but it isn't a Blomley. He wanted to prevent class AB in his design and have two independent sub-amplifiers as the outputs.

 

[dreamth]

2n3904 is a 300 Mhz device while 2sc1845 is a 110Mhz device...how much higher do you need than that and WHY?
And for the cascode transistor in general beta is irrelevant.You just need to give it enough base current.Usually the higher the beta , the lower the FT...

 

[Alexandre]

The reason why the splitters have to be as fast as possible is that their "bias point" is very very low in current, and they switch.
2n3904 and complement is the suggested device in Blomley's article and builders had good results with it. It doesn't hurt to have a couple of alternative devices to use in that location.
Beta is relevant because its nonidealities make nonlinear distortion. This distortion will be attenuated by feedback but you can't say beta is irrelevant, even in a cascode.

 

[dreamth]

The reason why the splitters have to be as fast as possible is that their "bias point" is very very low in current, and they switch.
2n3904 and complement is the suggested device in Blomley's article and builders had good results with it. It doesn't hurt to have a couple of alternative devices to use in that location.
Beta is relevant because its nonidealities make nonlinear distortion. This distortion will be attenuated by feedback but you can't say beta is irrelevant, even in a cascode.

Actually I can say that...the problem I have with such vague statements that sound like the theory of everything saying particularly nothing about anything is that I doubt Roger Penrose gave Hawking a PHD title on empty words with no formula attached to it. If there was no Bloomley article with a real incarnation of his amplifier attached to it you wouldn't have any transistor swap in mind would you? I was as doubtful as apostle Thomas on 1Mhz germanium transistors capabilities until I saw this video and that cured me completely of this FT nonsense thrown at everything without discrimination:

Yes, that is 100kHz sinus obtained with 1MHz trz! That is because 1MHz is 10 x 100kHz not because pigs are flying higher than pigeons.Pigs do jump higher than pigeons, but they don't fly...

 

[Alexandre]

If beta was irrelevant the bjt cascode would be ideal! Why bother then with tube cascodes, or unobtanium mosfets such as the k214/j77 ?
The answer is beta is important. It varies with ic, vce, junction temperature. Higher beta in a cascode = less distortion.
Nice youtube channel.

 

[Alexandre]

FT nonsense thrown at everything without discrimination

Not the case here. Maybe you didn't understand the circuit: the "cascodes" are almost turned off (very low current, under 1uA) at the zero crossing of the signal. At certain levels (close to zero crossing) they switch. One hands over to the other. High ft correlates with fast switching, it's an indication that the part might work well in this circuit (introduce minimum distortion). And it's nice to try different parts and measure/listen to the differences if any.