| Ultima Thule |
Hi,
after a small chat in Suzy thread, ETI 5000 MOSFET Power amp, I posted a link to a schematic from Dr Bora on his Delta amplifier which is quite unusual at least in that sense it uses in audio gear a very seldom seen "Cross Quad" diff pair at the IP.
That led me to start this thread as I wanted to know if other have seen and/or used it, and bring it out in the light.
At a first look it cheats one's eyes as it doesn't behave as "expected" with a normal diff pair, it has actually an inverted output in comparison to the "normal" diff pair.
Analog Devices shows an exampel of the cross quad in their datasheet for
MAT04 which starts on page 8 with a shematic seen on p.9, though with an additional diff pair on top which in this case acts as the input.
Dr Bora's Delta amplifier schematic can be seen here.
Your thought's and experiences?
Cheers Michael |
|
|
| ilimzn |
| quote: | Originally posted by Ultima Thule
At a first look it cheats one's eyes as it doesn't behave as "expected" with a normal diff pair, it has actually an inverted output in comparison to the "normal" diff pair.
|
Not as far as I can see in the Delta schematic...
It should also work with MOSFETs, but most simulators have problems with it, sop i guess it's back to the old 'build it and see' ways :) |
|
|
| darkfenriz |
Ask directly dr Jagodic whether he built it and if so, how does (n't) it work.
I'm interested....
About croos-quad... I have not used it, I don't see many advanteges, but one huge disadvantage- possible saturation of some of input bjts at clipping/slew overload.
Look to Hawkford's site for more info.
regards |
|
|
| Ultima Thule |
Darkfenriz,
you should now the answer yourself! :D
I should point out that it's my belief too that the Delta schematic is drawn wrong, you made the right correction Darkfenriz.
Did you noticed that Ilmzin? :cool:
And yes, I have read the Hawksford paper.
Cheers Michael |
|
|
| darkfenriz |
The strange thing is... I started to doubt.... and simmed just as author drawn it and it simmed (1kHz) flawlessly with both 'positive' and 'negative' feedback...
unless fed with a sqare, than the original design showed oscillations at some 8MHz
Great mistery to me
I think this can have something to do with saturation of cross-quad and changed polarity of inputs (TL071 style), but I don't know.
The only 'rational' explaination is the simulator gets fooled by kind of classD work of some stages with high carrier frequency... and it is not visible in sim.
So this actually could work, that's why I would be interested.
very interested
regards
Adam |
|
|
| Ultima Thule |
Ok,
I see...
Because when I look at the cross quad I do understand why it get's inverted at it's output, hmm.. does it behave differently for small signals compared to large signals?
Adam, did you sim just the cross quad alone, it would be best to just isolate it for now and study it, if I find the time I will build it tomorrow Friday.
Thanks for your input,
Cheers Michael |
|
|
| darkfenriz |
| quote: | Originally posted by Ultima Thule
does it behave differently for small signals compared to large signals?
|
Exactly
horisontal is difference voltage and the curve is collector current of one output 'legs'.
So sometimes it can work provided never switched to different 'monotonics' part of the curve, which can happen only in noiseless/clipless/TIMless simulators.
I guess it deserves further investigation.
cheers |
|
|
| davidsrsb |
I simulated the cross quad against a conventional LTP at the same current, collector load etc.
Phase is reversed
Gain is about 30dB greater
Gain Bandwidth Product is the same |
|
|
| davidsrsb |
| It needs emitter degeneration to stop it oscillating on sim. I just added some 10Rs and now the bandwidth is improved, but becomes 2nd order at roll off. Gain is still much higher than the simple LTP |
|
|
| Tim__x |
How does it work? :xeye: My eye sees a positive feeback loop around Q3 and Q4, actually I see positive feedback into both the base and emitter of both Q3 and Q4.
I tried simming it, it works fine in small signal analysis, but in .tran it quickly latches up and acts like a normal LTP. |
|
|
| mirlo |
| What's the claimed advantage of the crossquad? Is it supposed to linearize the diffpair? The gain improvement can be acheved with a simple cascode ... |
|
|
| davidsrsb |
The circuit seems very touchy . Q1 is working as an emitter follower. Q2 is the gain stage and Q3 runs in coomon base.
As Q2 turns on more. Q4 turns off and Q1 also acts as a common base stage. This is why the current in Q1 decreases as the base goes more positive. So this is positive feedback but not enough to oscillate.
The circuit simulates as being much more linear than the LTP with far more abrupt limiting |
|
|
| Eva |
The cross quad LTP is just a low part count version of this, I think:
It's a darlington LTP with the inner transistors cascoded to operate at a fixed 0.7V Vce. However, the collectors of the outer transistors are cross-connected to the load resistors and this ruins the gain and phase characteristics at high frequencies because the base currents of the inner transistors contribute to the wrong output leg (a plain darlington would behave better).
I don't see any advantage in that circuit other than part count reduction, because the higher gain, the 2nd order roll-off and the cumbersome latch-up are absolutely undesirable characteristics. We all have been adding degeneration resistors to our LTPs for a long time now in order to reduce their gain, so why would it seem attractive to add 30dB of gain now? |
|
|
| davidsrsb |
Not really
At audio frequencies the input transistors collector currents are nearly constant and modulating the bases of the output transistors has little effect. |
|
|
| mirlo |
It is different from the darlington LTP because the emitter follower/cascode that drives the base of the other side, carries current of the opposite side driving it. So it has Vbe that varies opposite to the Vbe variation of the diffpair transistors.
In other words the Vbe's of the emitter followers adjust slightly to predistort the signal that goes in to the emitter coupled pair. This enhances linearity.
It seems like a really elegant solution, would work with enhancement mode MOSFETs just as well, to reduce input stage nonlinearity, and MOSFETs wouldn't suffer the saturation/reverse operation problem when overdriven that can cause latchup.
:cool: |
|
|
| ilimzn |
| quote: | Originally posted by mirlo
It seems like a really elegant solution, would work with enhancement mode MOSFETs just as well, to reduce input stage nonlinearity, and MOSFETs wouldn't suffer the saturation/reverse operation problem when overdriven that can cause latchup.
:cool: |
That was exactly what I wrote about in my post. Unfortunately, simulating it with the MOSFET models I have at hand is even more touchy than with BJTs. Also, if you use all the same MOSFETs in tbe simulator (impossible in real life, of course), the simulated distortion goes down to the order of the sim numeric error :) |
|
|
| Eva |
| quote: | Originally posted by davidsrsb
Not really
At audio frequencies the input transistors collector currents are nearly constant and modulating the bases of the output transistors has little effect. |
At RF nothing is constant and base currents are no longer negligible in comparison with collector currents. Good routing of base currents makes the defference between poor and good stability.
| quote: | Originally posted by mirlo
It is different from the darlington LTP because the emitter follower/cascode that drives the base of the other side, carries current of the opposite side driving it. So it has Vbe that varies opposite to the Vbe variation of the diffpair transistors.
In other words the Vbe's of the emitter followers adjust slightly to predistort the signal that goes in to the emitter coupled pair. This enhances linearity.
It seems like a really elegant solution, would work with enhancement mode MOSFETs just as well, to reduce input stage nonlinearity, and MOSFETs wouldn't suffer the saturation/reverse operation problem when overdriven that can cause latchup.
:cool: |
MOSFETs would require both transconductance and Vgs matching in 4 devices, this is not a trivial task at all.
I think darlington-like approach still wins:
This works the same or better provided that the model of PNP and NPN devices is chosen to have similar transconductance around 1mA. Furthermore, the current mirrors with high value resistors work only for audio frequencies and roll off quickly at RF, thus improving phase margin. As a bonus, this circuit should not suffer from latching or phase reversal issues, altough turn-on after a overdrive period relies on leakage currents and positive feedback, so it may probably take some hundreds of nanoseconds. Two additional 1uA current sources may be added to ensure class A operation of the PNP devices and proper recovery after overdrive.
Note that ultra-precise current sources are not required, because what matters here are just current slopes.
|
|
|
| Eva |
A funnier version:
It should be also latching-free and phase-reversal free. The four lower transistors and their emitter resistors would require some matching for best results, though. |
|
|
| mirlo |
For the version with the PNP emitter followers, it looks like you might need another output leg on the current mirrors on top to actually have an output (?)
I have a bad habit of presuming matching is no problem.
Does anyone make monolithic MOSFET arrays? Like MAT-04 except MOS?
|
|
|
| mirlo |
| Oh, silly me, I see you are using the current mirror degen resistors on top to double as load resistors. Cool. |
|
|
| Terry Demol |
Hi all,
I have the cross coupled pair sussed out completely. Sorry, have
to go to bed will reply in morning, it's been a big week.
Cheers,
Terry |
|
|
| Eva |
Humm...
From now on, I think that I'm going to use the last circuit in all my projects. It has another nice feature that makes it superior to the 4 transistor version: The Vbe corrected transistors are operated all at a similar Vce. This is a great thing because d(transconductance)/d(Ic) "slope" changes a lot with Vce, and a transistor operating at 1.4V Vce is not going to match at all with another one operating at 40V or so. |
|
|
| Elso Kwak |
| quote: | Originally posted by Terry Demol
where |
further on in that thread |
|
|
| darkfenriz |
| quote: | Originally posted by Eva
This works the same or better provided that the model of PNP and NPN devices is chosen to have similar transconductance around 1mA. Furthermore, the current mirrors with high value resistors work only for audio frequencies and roll off quickly at RF, thus improving phase margin. As a bonus, this circuit should not suffer from latching or phase reversal issues, altough turn-on after a overdrive period relies on leakage currents and positive feedback, so it may probably take some hundreds of nanoseconds.
|
Yes, after any type of overdrive it can severely oscillate, therefore this a nice tim generator, or be positively fed back (lack current thru one leg implies switching off of its 'driver') so that everything turns into smoke. |
|
|
| darkfenriz |
| quote: | Originally posted by Eva
A funnier version:
It should be also latching-free and phase-reversal free. |
Are you sure? |
|
|
| Eva |
| quote: | Originally posted by darkfenriz
Are you sure? |
Try it, it simulates fine. According to Pspice, it behaves well no matter how much overdrive is applied, as opposed to the original cross quad that latches badly.
Also, according to Pspice, it has slightly better phase margin than the original circuit, that has a peak at 22Mhz (BC556B) that never disappears (those naughty base currents badly routed), no matter how much compensation is added (in all circuits that I tried Pspice requires 10pf capacitors between each input and output node for stability). THD and gain figures are similar.
|
|
|
| davidsrsb |
| quote: | Originally posted by darkfenriz
Are you sure? |
Behaves OK on my sim, but what is the advantage? The gain is just x10 and 20MHz -3dB
One thing about the basic crossed quad is relatively poor supply rejection. Further cascoding is needed, but we probably need a higher voltage stage anyway. |
|
|
| Eva |
| The advantage is not the gain but the Vbe linearisation. Anyway, classic amplifier circuits have no room for more gain. |
|
|
| mcp |
| quote: | Originally posted by davidsrsb
but what is the advantage? |
The advantage could be improved CMRR. |
|
|
| Elso Kwak |
The advantage is how it sounds...
Now how do you SIM that???:cool: |
|
|
| forr |
The cross quad is related to the name of Caprio.
It was one of the very linear differential circuits studied by Hephaistos-Gérard Perrot in the magazine L'Audiophile, back in 1988. The conclusions were that :
- the linearity is dependant on the good matching of the devices
- it is not a very stable circuit with a possible negative input impedance
- it does not behave well at start-up or when overloaded.
In the same magazine, Hephaistos had a look on the Van De Plaasche's circuit, some of the schematics shown in this thread should be apparented to it.
~~~~~~ Forr
§§§ |
|
|
| Terry Demol |
| image insert test |
|
|
| Terry Demol |
| I'll try a bigger image this time |
|
|
| Terry Demol |
Hi Guys,
Hopefully I have got this right :)
I have analysed the xcoupled quad as a simple open loop dif pair
with degeneration (Rdeg).
Each IP has a 0.1V step. So the total voltage step between + and - IP's = 0.2V.
Before we get too far remember that the XCQ inverts polarity so
as Q1 goes + its current decreases and as Q2 goes - it's current
increases. You have to visualise this clearly or everything after gets
confusing.
I have added an arbitrary delta BE for each transistor which
basically represents the inherent non linearity of a BJT's B-E
voltage versus current.
So as Q1 goes 0.1V + its delta VBE is 30mV, due to decreasing
current, which is comparatively larger than Q2's delta VBE of 10mV.
As we all know this is due to Q1's emmitter resistance increasing
and Q2's emmitter resistance decreasing.
What the XCQ does is transfer the delta VBE from Q1 to Q4 and
from Q2 to Q3. This has the resultant effect of linearising the
resultant current flow through the dif pair.
As can be seen the resultant total emmitter swing of Q3 and
Q4 is exactly the same as the IP swing.
IP swing = 0.1 + 0.1 = 0.2V
Q3/Q4 emmitter swing = 0.12 + 0.08 = 0.2V
So the current through the dif pair is completely linearised.
As an additional note, since the XCQ reverses phase, as Q1
swings + and its current is reducing, its miller c is now acting
as a small + feedback mechanism as opposed to the normal
- feedback. This is probably one of the reasons for increased
susceptibility to oscillation.
As another additional note, it can be seen that since the
emmitter swing of Q3 and Q4 are now the same total resultant
swing of Q1 and Q2 bases, ie; the inputs, these emmitters
constitute a zero impedance.
This is probably what is throwing the sims out ans also a reason
that this arrangement MUST be degenerated and probably with
quite a significant R.
Note I have re attached image as original has a few mistakes.
I hope I have not been too confusing.
Cheers
Terry
PS, please be constructive with your replies, this took a bit of time
to nut out and draw. |
|
|
| Terry Demol |
| revised image please disregard previus 2 images |
|
|
| davidsrsb |
| Adding a cascode section makes the original circuit much more stable. Not really suprising that having a large in phase signal on the collector of the input transistor can cause stability problems. |
|
|
| Terry Demol |
| quote: | Originally posted by davidsrsb
Adding a cascode section makes the original circuit much more stable. Not really suprising that having a large in phase signal on the collector of the input transistor can cause stability problems. |
Did you check the sim with a) no cascode b) no 100R IP resistors?
What are the response differences? |
|
|
| davidsrsb |
The input resistors have an effect
Green trace is 0R1, Red is 100R |
|
|
| darkfenriz |
Eva
Yes, I got it! some of bjt's do saturate at overdrive, but this does not affect output.
Good job.
This can be useful in projects with differential stage having huge effect on overal harmonic distortion.
regards |
|
|
| Eva |
I have the 6 transistor version working in breadboard now, and indeed it works. Effectively, it has negative input impedance, and this necessarily causes oscillation at HF. My prototype is made with On-Semi BC556B (these feature very low Vbe change with Ic) and oscillated at 16Mhz like crazy until I added 22pF miller capacitors to make the input impedance back positive at HF. I'm also considering resistive impedance cancellation.
The real prototype confirmed that my version does not latch up and does not suffer from phase reversal when overdriven.
I'm using this input stage in a N-channel rail to rail project where the output HEXFETs are part of a CFP. Such a circuit is very hard to stabilise, but in return for the effort it does not require any thermal compensation, it does not suffer undesired cross-conduction, and the chore of matching output devices is avoided by using one CFP cell for each one. I'm going to open a thread to show it one of these days since I've just got to work a preliminary version with +-18V rails and one pair of IRF640 :)
|
|
|
| Elso Kwak |
| quote: | Originally posted by Eva
I have the 6 transistor version working in breadboard now, and indeed it works. Effectively, it has negative input impedance, and this necessarily causes oscillation at HF. My prototype is made with On-Semi BC556B (these feature very low Vbe change with Ic) and oscillated at 16Mhz like crazy until I added 22pF miller capacitors to make the input impedance back positive at HF. I'm also considering resistive impedance cancellation.
The real prototype confirmed that my version does not latch up and does not suffer from phase reversal when overdriven.
I'm using this input stage in a N-channel rail to rail project where the output HEXFETs are part of a CFP. Such a circuit is very hard to stabilise, but in return for the effort it does not require any thermal compensation, it does not suffer undesired cross-conduction, and the chore of matching output devices is avoided by using one CFP cell for each one. I'm going to open a thread to show it one of these days since I've just got to work a preliminary version with +-18V rails and one pair of IRF640 :)
|
Why not begin with building the circuit in the MAT04 datasheet and go from there???? BC556B not matched transistors
:confused: :confused: |
|
|
| Eva |
| Of course these are not matched, but I got 100 BC546B and 100 BC556B (both On-Semi) for less than 10 euro :) Do you know how much does a MAT04 cost? ;) |
|
|
| Elso Kwak |
| quote: | Originally posted by Eva
Of course these are not matched, but I got 100 BC546B and 100 BC556B (both On-Semi) for less than 10 euro :) Do you know how much does a MAT04 cost? ;) |
Ever heard of samples Eva? Are you a woman?
:rolleyes: |
|
|
| Terry Demol |
| quote: | Originally posted by Eva
I have the 6 transistor version working in breadboard now, and indeed it works. Effectively, it has negative input impedance, and this necessarily causes oscillation at HF. My prototype is made with On-Semi BC556B (these feature very low Vbe change with Ic) and oscillated at 16Mhz like crazy until I added 22pF miller capacitors to make the input impedance back positive at HF. I'm also considering resistive impedance cancellation.
The real prototype confirmed that my version does not latch up and does not suffer from phase reversal when overdriven.
I'm using this input stage in a N-channel rail to rail project where the output HEXFETs are part of a CFP. Such a circuit is very hard to stabilise, but in return for the effort it does not require any thermal compensation, it does not suffer undesired cross-conduction, and the chore of matching output devices is avoided by using one CFP cell for each one. I'm going to open a thread to show it one of these days since I've just got to work a preliminary version with +-18V rails and one pair of IRF640 :)
|
Eva,
The original croos coupled circuit has fewer components and
theoreticall perfect linearisation of VBE voltage, especially when
used with an almost ideal and perfect matched device such as
MAT04.
Non of these will have that degree of inearisation, however I can
see #2 (post 17) has the advantage that it can swing significant
IP voltage. I like this one and have experimented with it myself.
However I found that at the end of the day the non linearity
due to VBE is only 1 factor amongst many that add up to
front end linearity.
#3 appears to me to be pretty much identical to Hawksford IP
arrangement from paper "distortion correction circuits for audio
amplifiers'- see fig 9, and cannot swing significant IP voltage
so the cross couple pair would be superior if implemented
correctly.
But as always YMMV so have fun.
I'd be interested to see how all these circuits measure in real life,
sims tell little to me, and how theory translates to practice.
Cheers,
Terry |
|
|
| AKSA |
Stability might be improved if the two cross coupled bases were interconnected with three resistors each of 1K.
Cross coupling would then be between resistors 1/2 and 2/3. This dilutes the level of crosscoupling , molifying the extreme effects.
Cheers,
Hugh |
|
|
| Eva |
According to the datasheet, for Ic=1mA my BC546B show unity gain at 100Mhz. On the other hand, a 22pF capacitor shows approx 68ohms impedance at 100Mhz (the value of emitter resistors that I'm using). So I tried a 22pF between both inputs of the 6 transistor version of the quad LTP, and it eliminated all the oscillation.
I think that compensating it this way is much better than using miller capacitors. Furthermore, in my amplifier the negative rail is bootstrapped to ensure rail to rail output, and this was causing me trouble because the 22pF miller capacitors that I was previously using had to be necessarily tied to that negative rail. The circuit turned unstable everytime the negative rail was bootstrapped and followed the output of the amplifier. The new compensation solved it.
Also, I've placed a 15K resistor between both inputs of the LTP. Considering that my transistors have a typical (measured) hFE of 250, and that 68*250=17000, that resistor should produce a good degree of cancellation of the otherwise negative input currents in the audio band, thus making input current almost constant and input impedance huge. |
|
|
| mirlo |
This is getting interesting. I hope you will post a new schematic when you are finished making adjustments, Eva ...
I guess a next step might be to try to invent a similarly absurdly linear VAS or perhaps output buffer stage. |
|
|
|
