TGM 1i - an integrated hybrid amp inspired by Hugh Dean

[Bigun]

Correct Christian, Hugh never felt that speaker protection was needed on AKSA. In fact he published somewhere a comment that it was designed to fail-safe, the one rail most likely going down first and leaving the feedback network intact to minimize the dc at the output - plus he had never heard or seen any of his AKSA amplifiers failing and producing a problem for the speaker. I think for DIY purposes we are all free to leave out such protection too since we are not selling products that might damage customers property if the customer does something stupid like short the output and blow the power devices. My desire for the output protection is more of a mute function - admittedly it could be placed somewhere earlier on the signal path but at the output it will serve double duty as protection. My plan is to review the circuit from TGM8, it does a good job with few parts and I will look at the feasibility of including footprint for both mechanical relay and SMD based FET relay. Putting FETs in power rails doesn't provide a mute function.

Hi Xrk971 - those opto-isolators work very well indeed. I used one in my TGM8 amplifier but I am looking to avoid the use of SMD parts for this build - keeping it 'traditional'. However, as mentioned above, I may be able to include footprint for both options.

In terms of the AKSA circuit, it has always bothered me (probably for aesthetic reasons only) that the RC filter is on the +ve rail where the current sources are arranged, when in actual fact it would do more good on the -ve rail wedged between the driver and VAS transistor, assuming of course that the LTP tail had a proper current source. But Hugh designed this to sound good ...

That's right, the AKSA circuit 'sound' depends on quite a few factors that are not that obvious and the filtering on the +ve rail is part of the total picture. Hugh posted somewhere that he had tried putting the diode-resistor-capacitor string in the -ve rail but for this particular amplifier it offered no sonic benefit. He did use current sources on the LTP for later designs but again that would no longer be the original AKSA in terms of sonics. And much later he abandoned the common LTP altogether. So for this project, we are trying to stay true to the AKSA sound, warts and all, a classic in it's own time.

 

[xrk971]

Bigun,
Could a DIP part like this work? They are like $1 ea.

https://www.fairchildsemi.com/datasheets/MC/MCT6.pdf

 

[PaulBysouth]

Bigun,
I've sent you a PM.
Regards,
Paul Bysouth

 

[Ranchu32]

So for this project, we are trying to stay true to the AKSA sound... a classic in it's own time.

Putting FETs in power rails doesn't provide a mute function.

Well, it can, provided you have separate unswitched (and in my case boosted voltage) rails to the front-end, which itself is able to control the feedback loop (albeit via a high impedance return path) while the output stage is disconnected, but I digress....

That's right, the AKSA circuit 'sound' depends on quite a few factors that are not that obvious and the filtering on the +ve rail is part of the total picture. Hugh posted somewhere that he had tried putting the diode-resistor-capacitor string in the -ve rail but for this particular amplifier it offered no sonic benefit. He did use current sources on the LTP for later designs but again that would no longer be the original AKSA in terms of sonics. And much later he abandoned the common LTP altogether. So for this project, we are trying to stay true to the AKSA sound, warts and all, a classic in it's own time.

I think that there are a lot of subtleties here that contribute to the good sound, that aren't apparent just taking the circuit at face value. On the subject of the LTP CCS, it has been my experience that an active current source in this position helps the circuit start up faster and thus minimises turn on noise. But I've never really spent much time with the AKSA-55 and I can't say how it fares in this regard.

 

[AKSA]

The AKSA was, in my view, a unlikely classic. This might be overstating it, but it did hit on a good midrange and top end. If you change anything it will sound different - not necessarily bad, but certainly different. I never had a lot of money for instruments, and figured that with my detective skills and left field thinking I might do something better. In the beginning, I had so little confidence in my thoughts that Richard built it first; he was more confident that I was, and perhaps more energetic. I owe him a debt of gratitude!

In the following years I came to see that many designers were blindly following their instruments, supplied by a cashed-up corporation no doubt. I was struck by the genesis of the AP1, which ushered in a new era of terrible sound and wonderful, accurate measurements. After this revelation, I resolved never to buy an AP. But I become very good at using a used, 25MHz CRO.

My next work was informed by my critical listening, and that of many good friends who happily listen to Coltrane, The Levy Bank, Jazz at the Pawnshop, Patricia Barber etc..... and gave me their thoughts. I took a few bad turns in all this, producing amps which did not sell well, but the AKSA sold because of the DIY marketing, the low AUD, and a huge email backend from me. This could not happen today; the eBay phenomenon and cheap Chinese products have killed this, and so to compete we have to field clearly superior products to win the customer.

However, after the Maya, the flagship, I am revisiting the AKSA just now. The work is done, the prototype is tested, and I'm confronting the assembly bill for the first batch. It will be called the Super AKSA, SAKSA to you and me, which it is an extraordinary amplifier and sounds absolutely entrancing. It will supplant the NAKSA 80, which is a very good amp, and is now deserving of retirement (like me, actually!).

The footprint will be the same as the NAKSA 80, and same price, $AUD930.

This might be interesting for you, Gareth! I've even gone back to a LTP!!

Thank you, and Greg Erskine, for your work on the AKSA. It is memory lane for me, and a classic for many people, and hopefully a very good amp for the builders. More than anything, it is humbling for me. Thank you Sirs.

Cheers,

Hugh

 

[Bigun]

I've even gone back to a LTP!!

 

[Bigun]

The AKSA was, in my view, a unlikely classic. This might be overstating it, but it did hit on a good midrange and top end. If you change anything it will sound different - not necessarily bad, but certainly different.

I may include footprint on the pcb for a revised VAS. I don't want to mess with the LTP, there are many trade-offs and current sources and current mirrors would make it a very different amp. But I believe perception of bass is limited by higher H2 that the base AKSA uses and this could be tightened up if the differential signal at the LTP were reduced - which requires a bit more OLG - and a revised VAS would provide a bit more OLG as well as extended bandwidth if wanted. It would still be a different amp, but retaining much of the AKSA flavour and sometimes it's just nice to have the option on-board. The layout is going to be tight though, something may yet have to be ejected from the list of desires.



I look forward to learning more about your new amp - which I expect will have big MOSFET outputs, single pair since you can now get lots of power without the expense of device matching.

 

[PaulBysouth]

Amp protection circuit

Gareth,
Attached is a design for an amp protection circuit that uses a speaker protection `relay' (actually two mosfets acting like a relay) to provide:
- A turn on delay,
- A mute function via a front panel switch,
- Fuse monitors (if either fuse blows the speaker is disconnected), and a
- DC offset detector.

This circuit could go on the amplifier PCB or on a separate PC. A simplified version of this circuit is used by Hugh in his Maya amp, where it works very well (as does the rest of the Maya amp - it's very good).

I've drawn the circuit to suit SMD parts, but could be simplified if you use through hole parts (eg combine R49 & R50).

The `speaker relay' is provided by m4&m5 which are turned on by the vom1271 opto-voltaic driver, which includes a fast turn off circuit. Connect a switch to short out pins 1&2 to get the mute function. The vom1271 needs 20mA to provide a decent drive, and this is provided by m3 configured as a current source. I've used a NDF04n60z mosfet, as this is the cheapest insulated To220 with protection zeners I could find, but any To220 hexfet would do (use a heatsink).

DC detect is provided by R33,c20 and the ps2506 opto-darlington. Turn on delay is set by R44&c21. I've used 15v zeners, but larger can be used (24v would probably be better, Hugh used 33v). If you want a front panel indication, then R40,R41,led1 can be used, else just delete.

If either fuse blows, then the m3 current source is turned off and hence the speakers are disconnected. NOTE: The fuse monitor etc circuit must be powered from the power rails AFTER the fuses.

For the `speaker relay' I've chosen FDP083n15a mosfets as they have very low RDSon (8mohm) and sufficient voltage (150v - needs to be more that the rail to rail voltage), but the irfi4228pbf is also a good choice (12mohm, but a lot cheaper). Hugh used the irfi4228pbf. T1 is optional (but cheap), and is used to protect the mosfets if a large external voltage was applied to the output

Regards,
Paul Bysouth

 

[Bigun]

That's an elegant circuit Paul !

Why does the max voltage rating for the 'relay' FETs need to be more than the rail-to-rail ? - as far as I can see the worse case is when the 'relay' is open circuit in which case the maximum it will see is one rail. When the 'relay' is closed circuit there won't be any appreciable voltage across it. Seems to me that a 60V rated part would be sufficient - granted that higher voltages are readily available but I want to understand your reasoning.

 

[PaulBysouth]

The reasoning for using `relay' mosfets with a voltage rating of twice the rail voltage is:

- Assume the amplifier is driving a bass speaker such that the cone is moving (maximum excursion type movement), and the `relay' opens at this time,
- the speaker will produce a back emf (coil moving in a magnet).
- assume that as soon as the `relay' opens the amplifier goes hard to the opposite voltage.
- the relay now sees the rail voltage (say 50v) plus the emf voltage (unknown value).

I think it is unlikely that the emf voltage would ever get to the rail voltage, but probably safer to assume it will, so we would need 100v mosfets if 50v rails where used. The irfi4228 mosfet is one of the cheapest mosfets around, so might as well use it. The FDP083n15a is about 3 times the price (but I have some, so that's what I'll use). T1 (a SMAJ90ca, which acts like two 90v zeners in series) provides some backup protection, but unlikely to be needed.

Incidently the irfi4228 mosfet has an RDSon of 12mohm (so 24mohm total for the `relay'), but this is significantly less than the typical speaker cable resistance.

pb

 

[Bigun]

The reasoning for using `relay' mosfets with a voltage rating of twice the rail voltage is ... the emf voltage

I had been worried that the back-emf cold be even higher than the rail voltage when you break the circuit, good to know you think it would probably be less. I incorporated catch diodes on my TGM8 amplifier just in case.

What about putting a cap across the relay 'contacts' to limit voltage spikes across the FETs (combined with catch diodes) ?

 

[AKSA]

No requirement, Gareth. The surge diodes will take anything beyond 90V, and the spike is likely far less than this. This circuit from Paul has never been a problem on the Maya, which is rock solid. To see these two series mosfets in the output node is astonishing; there is absolutely no effect on the sound quality - seamless - and they completely protect any speaker.

Hugh

 

[Ranchu32]

Back to the AKSA circuit at #18 for a moment.

I can see two circuit elements that have almost certainly been chosen to shape the distortion profile:

1. The choice of R4 unbalances the input pair by varying degrees, and will promote H2 distortion. Any attempt to better balance the pair (either by "optimising" R4 and Q2's collector resistor, or inserting a current mirror) will undoubtedly change the sonic profile.

2. My suspicion (to be confirmed in Spice) is that the tail current is being modulated through the supply rail as the output stage draws varying currents. D1, R10, C3 and R3 are the critical components here and I suspect that any change here will also have an influence on the sound.

Another source of distortion (possibly unintended) is the modulation of Cdom via the negative rail. The reason I floated the idea of inserting an RC filter between the VAS and driver (say 22R and 1000uF) is that it would clean up the ripple entering Cdom without affecting LTP balance, hence harmonic profile (hopefully). It is also possible to achieve the same net result by cascoding the LTP and moving the Cdom connection, but that is more complex, and the more active devices enclosed within Cdom makes it more susceptible to instability.

I don't know if this has ever been tried with the AKSA. I did this experiment with the P3A (superficially similar front end topology) and it measured better, although admittedly, I couldn't say that I was able to identify an audible difference.

 

[Ranchu32]

So for this project, we are trying to stay true to the AKSA sound, warts and all, a classic in it's own time.

A couple of other points in relation to the AKSA sound.

I have a friend here in Brisbane who has owned, for a considerable number of years, an original AKSA-55, which features to this day in his main listening room set up.

We've put this AKSA up against a number of other DIY amps, including several of my own, and both agree that the midrange and top end is among the very best. His chief complaint (probably his only complaint) is that the bass is a little thin - I tend to agree.

I've had the unit in my workshop and can't fault it. The power supply is very robust and can't possibly be the cause. We've experimented with different input and feedback caps of various size and type, which has resulted in relatively minor effect on the sound - good and bad.

I've been wondering:

a) whether this has been a commonly reported observation from owners.
b) about the underlying cause assuming its not an isolated example. I'm going to speculate that it might have something to do with the simple resistor current source used to load the LTP tail, that heavy bass notes causes a voltage drop across the resistor load, reducing its output impedance and unsettling the input pair.

Bigun, the second point I want to make is that you seem to want degenerate the input pair to improve its linearity (or perhaps overload handing in the face of this design's moderate OLG). I think you also talked about elaborating the VAS, presumably to improve linearity of this stage or increase overall loop gain.

I wonder if you can have your cake and eat it to by increasing the tail current match the gain otherwise lost through degeneration. That way you might achieve your objective without introducing an extra gain stage with its added phase shift. The only fly in the ointment seems to be that LTP resistor load, which would need to be reduced in value to match the higher current requirement, probably resulting in more ripple break through and reduced OLG (if the LTP's current source has a low output impedance I think the stage gain of this stage will be compromised).

I can think of two options: one is to use an active current source, the other is to greatly increase the supply voltage (either to the amplifier as a whole or a dedicated subrail feeding the LTP tail resistor.

 

[Bigun]

the tail current is being modulated through the supply rail as the output stage draws varying currents.

Correct, it's actually related to the Gilbert Cell (apparently I was the 3rd person ever to recognize this mechanism was being exploited in Hugh's amplifier) and you can control the effect with various component choices, including an additional resistor (see my TGM thread).

Interestingly, a number of people who have played and tweaked this design on the bench have reported that better LTP balance results in slightly better sound. Back in the early days there was a strong following of tweakers and much has been reported on the internet if you look for it (check the Audiocircle forum).

a friend here in Brisbane ... chief complaint (probably his only complaint) is that the bass is a little thin ... The power supply is very robust and can't possibly be the cause... different input and feedback caps of various size and type, which has resulted in relatively minor effect on the sound

Yes, I think others have reported similarly, the bass extends plenty deep (i.e. the caps are sized appropriately) but it is not 'punchy' in the same way some amplifiers are

I've been wondering... it might have something to do with the simple resistor current source used to load the LTP tail

I believe the issue is related to low order harmonics, bass tends to get softer and more spongy under such circumstnaces. This is a common thing with tube amplifiers, and AKSA was designed to exploit this harmonic structure of falling harmonics (ala Hiraga) for it's good sound. Furthermore, the LTP performs poorly with large signal swings, especially without large degeneration (> 100 R). As I mentioned above, higher OLG would reduce the differential signal at the LTP and reduce the distortion, coupled with a compound VAS (the simple bootstrap loaded VAS is another source of low order harmonic distortion) and on-board reservoir capacitors, I feel this would tighten the bass. But that's not the AKSA amp. We know how to make a cleaner and more powerful sounding amplifier if we want to.

Bigun, the second point I want to make is that you seem to want degenerate the input pair

I'm proposing a low level of degeneration, this was part of the 'Nirvana' upgrade Hugh made to the amplifier (he used 10R I think). Listening tests showed a subtle improvement but this is very little degeneration, not the level normally used when talking about 'Blameless' designs. The simple LTP structure as used in AKSA can't be loaded with lot of degeneration or many of the competing tradeoffs around it start to fail -

I wonder if you can have your cake and eat it to by increasing the tail current match the gain otherwise lost through degeneration

In this simple architecture you can't do much about the tail current or you have to reduce R4 and then you lose OLG plus you change the relative modulation of the input pair. I've looked at this AKSA input stage from every direction over the years and many folk have tweaked it - unfortunately I don't think there's very much room to manouvre - the very first version of AKSA had very low tail current for example because of these tradeoffs.

In later designs Hugh changed the LTP and increased the tail current considerably. My TGM2 is based on his Lifeforce amplifier and it uses 4mA tail current. He then evolved several more designs with the LTP getting even more fancy with current mirrors etc. (I can't reveal too much more unless Hugh wishes to post the information). The only real difference between TGM1 and TGM2 was the LTP linearity and tail current - the bass response was clearly more punchy with the TGM2 and the sound had more clarity overall, but in my view it lost some of the treble 'magic' of the TGM1 (note, it is important to point out that my TGM2 is not an exact clone of the Lifeforce amplifier).

I believe there's a lot of subtle issues to understand and design around, which is why Hugh can pull off such amazing designs. The brute force approach that I took with TGM8 was to reduce distortion to tiny levels, with only 2nd and a bit if 3rd harmonic surviving at higher signal levels - and it sounds superb with delicate treble and dynamite bass, but of course, it's more complex and more difficult to build.

 

[Ranchu32]

In this simple architecture you can't do much about the tail current or you have to reduce R4 and then you lose OLG plus you change the relative modulation of the input pair. I've looked at this AKSA input stage from every direction over the years and many folk have tweaked it

You're right Bigun, I forgot that we are discussing a stage with a simple resistor collector load, not one loaded with a current mirror. That will teach me for posting late at night

 

[AKSA]

Bigun, Ranchu,

A few thoughts:
You are considering the effects on the resistor feed to the LTP from downstream.
Why not consider the input factors themselves? Why is the distortion even order?

Hugh

 

[Bigun]

A few thoughts:

it's never a few thoughts, there's a deep well of thoughts behind this one !

Not sure I understand where the Master is leading us on this. Of course the simple LTP we're looking at is not symmetrical enough to cancel 2nd harmonics because one 'leg' is providing current to the VAS and to Cdom, a frequency dependent non-linear load.

I have a question: why do people (including me) put an RC filter at the amplifier input to block r.f. ingress when it is usual practice in the r.f. 'world' to put such filters close to the entry point i.e. would it not be better to wire up an RC filter at the RCA connector to block the r.f. where comes into the chassis - where the parasitic signal current could flow back to the source through a cap soldered point-to-point on the back of the socket ???

 

[Bigun]

I decided to go with the solid state relay because it's so much better than most mechanical relays both sonically and in ability to shut down the output in the face of an inductive load with high dc current.

The circuit should be simple to minimize pcb space so I didn't attempt to incorporate all of Pauls' ideas this time around.

I've proven the circuit out on my TGM8 amplifier. It provides a turn-on delay (relatively short), quick disconnect when dc is present and a shut-down when it detects that the power rail is collapsing (i.e. amplifier turn-off).

 

[Bigun]

Not sure if this is very clear (the output devices are a faint dashed-line outline). It's first stab at a pcb for the amplifier. I'll be making something similar to my TGM8 - i.e. the power devices go underneath the board (the Vbe multiplier device is sandwiched between one of the power devices and the underside of the board) and then you bolt through the board and power devices into the heatsink so as to hold the whole shebang in place (just like Hugh's latest amplifiers). No other mounting hardware needed.

It's about 100mm x 70mm. The only SMD part is the opto-driver for the solid-state relay. You can get through hole versions but since I have some of the SMD parts in my parts bin I'm staying with those. Everything else is through-hole of course. Compared with the original AKSA board I have added the dc-protection circuit, an output inductor, a potentiometer to adjust dc-offset and an rf-filter on the input. Otherwise it's more or less the same.

I've tried to keep large AC current paths short and mostly in the lower half of the board. The drivers and VAS are in the top-left, the dc-protection on the right and the input diff-amp in the quiet top-right corner area. The SS relay (pair of TO-220) FETs are standing proud in the centre of the board.