At a practical level I would agree with you. My experience is that below about 0.1% you are not going to hear anything that would allow you to pick out a 1 ppm design over a 100 ppm design and I have to side with Nelson Pass on this one.
But, engineers being what they are, low distortion designs are a challenge and its fun.
I try to keep a balance, and put as much effort into the physical design and construction. I get little pleasure out of sims- they are simply a means to an end with me.
In my view, ANY topology can be made to sound good or bad, using any device you care to name (BJT, FET, MOSFET, IGBT, tube, etc).
Again, in my view, the trick is to use such a topology as will yield least distortion products in open loop, allowing us to use less global NFB for quality sound and uncompromised stability.
Not easy, but can be done. Case in point - Harman/Kardon Citation 24 power amp. Nominally 100W/8 Ohms, two pairs of output devices, solid dial mono type (transformer has joint primaries but separate secondary widings for each channel) PSU, uses some 12 dB of global NFB, but obviously lots of local NFB. Fully complementary.
It manages what I see as the ultimate trick, which is to make itself disappear. It is one of the most neutral sounding amps I have ever ancountered since 1970, nothing pushes, nothing pulls. Yet it has oodles of power on tap and is not one bit bothered by working into known diffiocult loads (e.g. AR94, the original series). The paradox is that it works best with a Luxman preamp rather than its stablemate preamp, Citation 22.
Admittedly, I do own speakers (B&M Acoustics, a local brand, built around Son Audax prime choice drivers) which are, to the best of my knowledge, one of the easiest loads ever anywhere (nominally 8 Ohms, minimum 6.5 Ohms, worst case phase shift -25 deg., 92 dB/2.83V/1m) in dynamic speakers. Just about everything works at its best with them, including supermarket models. That always helps.
I will be leaving in about 30 minutes for my summer vacation in Greece, so any replies will have to wait until August 18. Sorry for the delay.
Again, in my view, the trick is to use such a topology as will yield least distortion products in open loop, allowing us to use less global NFB for quality sound and uncompromised stability.
Not easy, but can be done. Case in point - Harman/Kardon Citation 24 power amp. Nominally 100W/8 Ohms, two pairs of output devices, solid dial mono type (transformer has joint primaries but separate secondary widings for each channel) PSU, uses some 12 dB of global NFB, but obviously lots of local NFB. Fully complementary.
It manages what I see as the ultimate trick, which is to make itself disappear. It is one of the most neutral sounding amps I have ever ancountered since 1970, nothing pushes, nothing pulls. Yet it has oodles of power on tap and is not one bit bothered by working into known diffiocult loads (e.g. AR94, the original series). The paradox is that it works best with a Luxman preamp rather than its stablemate preamp, Citation 22.
Admittedly, I do own speakers (B&M Acoustics, a local brand, built around Son Audax prime choice drivers) which are, to the best of my knowledge, one of the easiest loads ever anywhere (nominally 8 Ohms, minimum 6.5 Ohms, worst case phase shift -25 deg., 92 dB/2.83V/1m) in dynamic speakers. Just about everything works at its best with them, including supermarket models. That always helps.
I will be leaving in about 30 minutes for my summer vacation in Greece, so any replies will have to wait until August 18. Sorry for the delay.
I believe the problem is not the speed of the input devices, but the parasitic capacitance of their bases in a LTP.
As the base parasitic capacitance in common emitter operation (VFB) is higher than the one of the emitter in common base operation (CFB) -reason why common base is often used in HF amplification-, you need to lower input impedance to reach the same cutoff frequency.
In a feedback loop, it is not the cutting frequency that matters, but the point where phase begin to turn in such a way that the differentiation will not be 0 anymore... a lot lower in frequency... a lot lower in impedance.
But, in a LTP, you need the two inputs to be the same impedance for 0 offset.
In the example i brought, the emitter resistance (input impedance of the feedback path) in the CFB version is 10 Ohms. Go figure !
Who want a power amp with such a very low input impedance ?
That is all the story about the CFB topology advantage (and the reason why some guy, one day, called this "Current feedback".
The study i had done with my crescendo can be done on most of the VFB amplifiers. It is not complicated, it cost only few resistances and the only difficult part will be to derate for stability.
A good way for everybody to make their mind about the musical qualities of the two topologies in REAL WORLD.
You know that mastering engineers used to record the peak levels bandwidth of each tune, before to apply their compressions curves for vinyls.
One of my best sounding tunes he asked-me for his demo tape (acoustic instruments) was... flat up to 16000.
Just the hit of kick drums ans cymbals are responsible of those very short high level in trebles.
Nice conclusion
Last edited:
But, I think the point the pro CFB people are making is that this is a very linear topology capable of performance that easily equals VFA
With a lot of efforts.
The necessary complementary pairs will never match as well as pairs of the same polarity which allow straightforward design and which use of can be restricted to the input stage.
Two very good explanations came forward . TID . Transient intermodulation distortion , jury is still out on that one . More or less I side with it if I have to . Then power delivery . An especailly horrid Sony amp claimed 170 W rms . It was gutless , I was obliged to have one in stock . Hi Fi choice measured it as 2 W 1R transient burst . The NAD 3020 192 W . Now that relates to what I heard .
The Sony TA 5650 of a few years previous was excellent . Sansui as far as I remember never made a bad amp . Kenwoood neither . Onkyo was what it always was , quirky yet nice . Always an Onkyo sound . Looking at the circuits never said why or how . Luxman I never understood , quirky like Onkyo . Different to any other . Yamaha might be an amp too good for me ( I admit to liking tube amps ) . Often when using them I got bored . The sounds was never less than excellent . I get bored in Mercedes cars so it's probably just me being a snob . I never heard a Sony Esprit amp . I just have a hunch I would love it .
When I asked how some Sony amps were good and others not it was suggested to me that the boss made the bad ones ! The good ones the young engineers who always moved up into video . Sony STRVF3 excellent , STRVF5 not so . I sold off the VF5 and some IMF TLS50 which were equally unloved . The couple just wanted a silly price for OK hi fi . They were very happy . A VF3 and some KEF Coda's would have been better . VF4 superb .
If you have an old Kenwood ( Trio ) give it the largest toroidal transformer you can of the same voltage . You will be surprised how good the power amp is . The money was saved in the PSU . Then upgrade caps .
I think these Crown amps were quite respected at the time and still are.
There is correlation between Harmonic Distorsion and Intermodulation.
And there is correlation between Transient Intermodulation (the protocols to measure are not straigthforward) and Harmonic Distorsion mesurements at 10 and 20 kHz which tell much more about Slew Induced Distorsion, which is all that matters towards transients as long as the amp does not clip.
What sort of things ?
Self has been very meticulous in studying thermal effects in output stages.
And all the tests I have done about dynamic thermal effects in input stages tend to prove they are not significant.
Maybe be not close enough.
Concerning caps, it is quite emblematic of differences in the approach of engineering that a major step was overcome by Cyril Bateman's study, only using harmonic distorsion measurements, mainly at 1 kHz :
Capacitor Sounds, Speaker Cables and Crossover Inductors.
A few time after publication, ceramic caps (provided they are of NPO/COG type) and bipolar electrolytics changed their status from horrible to very desirable components.
Theses are great exemples as well and there s many more of thoses
past products that are hardly rivaled by recents products set apart
in the extraordinary claims department , not in perfs of course , but
with the senseless "good sounding" ubiquitous pseudo argument that
ended branding thoses fines amps as sometime "inferior" to ultra non
linear SET amps and other NFB less dubious "designs" when it is not
pseudo CFB ultra hyped very average amps.
When a book title contains words "audio power amplifiers" and there is not even mention of CFB topology, isn't it like foisting VFB to audio comunity? If CFB is bad or completely unsuited for audio power amps, the authors should write that in the book, give arguments for such opinion and let audio community judge the value of that statement.
I never questioned the competency and importance of authors like Doug or Bob, in fact I read their articles and books with great interest. There is no need to distort the meanings of my or any others posts.
Last edited:
See (short) mention of CFB ed #6, pp213-215 under the name of
"The Series Input Stage Push-pull VAS".
Neither of the two serious books currently available on power amp design, albeit very informative and full of facts, pretend to be of encyclopedic coverage.
Maybe some guys here could go in a partnership to write the missing book :
"Current Feedback Audio Power Amplifiers Design".
Then some other guys would certainly protest against such a mis-named title.
Who was the inventor of this name ?
Mark Alexander, with its quite original power amplifier ?
(The input stage was an OPA with feedback applied at its output. The push pull signals were kept from current sensors at its power supply rails.)
Not true forr.
Offset adjustment is easy, and as I remarked in my sx and nx-Amp write ups, the offset is remarkably consistent and low for such simple designs - they are fully DC coupled. The sx-Amp only has 28 dB of feedback and neither of these designs uses a servo.
But, why make this a point, when many VFA designs now days use a servo?
Last edited:
That is really the problem: reducing a big and important family of amps to a book footnote!
The consumers do not really have a chance to compare CFB and VFB because VFB dominates. If consumers were given a chance to compare, no doubt that many would choose CFB. And the reason that VFB dominate is because in industrial environment it is easier to mass produce VFB amps.
Every engineer is forced to be industry servant because every engineer needs salary. But when writing a book he can allow a luxury of being independent thinker!
Matching the complementary pairs will always be a problem but there are the THAT transistor packages that contain 2 PNPs and 2 NPNs that could be used for the front end of diamond input CFA. Cascoding could be used for higher voltages.
Even though the PNPs/NPNs do not totally match at least there would not be a problem with temperature matching.
Even though the PNPs/NPNs do not totally match at least there would not be a problem with temperature matching.
There is'nt - at switch on I measured 24mV on my sx-Amp and when its warmed up its about 1mV and very stable. The nx-Amp is similar.
Bottom line is the output offset issue on bipolar input power amplifiers pretty overblown. I think with JFET input designs that are direct coupled, a servo is mandatory.
Bottom line is the output offset issue on bipolar input power amplifiers pretty overblown. I think with JFET input designs that are direct coupled, a servo is mandatory.
mcd99uk,
no, devices were not matched. If you look at a bunch of 547/557, you will see that the Vbe is generally within 2-3mV of each other (I bought mine loose from Digikey about a year ago - Fairchild).
The offset adjustment is accomplished by injecting an offset current into the feedback summing node from a pot divider.
The transistor offsets are genrally well below the resistor tolerances of 1% - so they are a bigger problem. I have not done any detailed analysis, but have no reason to believe trimming the offset introduces significant amounts of distortion - its a technique used in most designs via a pot or servo amp.
Re the amps - all of my designs to date have been complimentary inputs.
no, devices were not matched. If you look at a bunch of 547/557, you will see that the Vbe is generally within 2-3mV of each other (I bought mine loose from Digikey about a year ago - Fairchild).
The offset adjustment is accomplished by injecting an offset current into the feedback summing node from a pot divider.
The transistor offsets are genrally well below the resistor tolerances of 1% - so they are a bigger problem. I have not done any detailed analysis, but have no reason to believe trimming the offset introduces significant amounts of distortion - its a technique used in most designs via a pot or servo amp.
Re the amps - all of my designs to date have been complimentary inputs.
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.