Aleph Mods

I've been asked, in another thread, to elaborate on my modifications to the Aleph style amp. So here is a new thread to jump-start the discussion.

As I mentioned in that thread, I modified the NP design and wound up with an amp that has 10 times less distortion, a freq. response to 200 kHz, and between 5 to 10 times greater S/N ratio (around -102 dB) than a stock Aleph. This actual noise content was about 100 microvolts; NP quotes 500 microvolts in his specs. I measured this directly: the first unit that I built, before further simulated optimization, measured a THD of around 0.04% THD at 1 kHz and 1 watt/8 ohm ouput. It produced around 20 watts class A with a +/- 22 volts power supply. Simulated distortion is now about 0.005 THD at 1kHz, and I presently building a unit based on this design using a double-sided circuit board that I may make available (presently the boards cost me about $100 apiece in quantities of 2).

As for the design, I popped in a current mirror transistor pair, in place of the 392 ohm resistor and the straight wire, that are in the drain legs of the input mosfet pair. This mod alone decreased distortions 10x. It also, if you've read Mr. Self's or Sloan's books, improves current matching in the differential input pair. This in turn improves the DC offset. My original DC offset, despite matching the input pair, was around 100 mV. Using the exact same mosfet input pair, the current mirror transistors decreased this to around 25 mV---and these were unmatched transistors. In my new amp, I am matching _everything_, including all transistors in all sections. By the way, I use ZTX455 rather than those specified by NP (mpsa18/ztx450). The ZTX455 have similar betas, but can handle almost double the voltage (they run around 50 cents each from DigiKey, so matching them up is easy and not too expensive).

I've also simulated (in Electronic Workbench) the design and optimized the IV control portions of the circuit so it can better deliver current into 4 and even 2 ohm loads, albeit with slightly more current and heat dissipation. I won't go into that discussion just now.

In addition, I designed the 90 x 205 mm circuit board to hold everything, including the power supply and up to 4 pairs of output mosfets: just plug in a transformer and the input and output jacks. This power supply is regulated: this is the only way to lower the noise level of the amplifier 5x over NP design. The cost factor is minimal: a couple of more mosfets and some diodes. In fact, the cost is probably less since you no longer need inductors (like Zen) or huge capacitor banks. I reduced my capacitors from 4 x 39,000 microfarads to 2 x 27,000 with 2 x 1800 microfarads following the regulator stage. It's basically derived from NP's regulator used in his preamp, but you need mosfets that can take several amps surge, so I selected different ones. I'll close here for now.

-Robert
 
Your results are very nice. Almost unbelieveable.
I am using PSpice for simulation. I put the Aleph 30 in it and ...THD at 1khz, 250mV input more than 2%! I made simulations of other amps (Hoffmann, Zen) and the results vere similar( sinus curve has deformated shape). I dont know were the mistake is - in design of the schematics or in the PSpice? More by the email.
 
Asen-

I went to a regulated supply because while the residual ripple was quite low (per NP's specs) the 120 Hz hum was quite audible with my speakers. Everything I build must be exceptionally quiet as my speakers (Lowthers) are horns with around 102 dB sensitivity. In other words, with my other speakers, which are about 86 dB sensitivity, there is only silence, so the residual hum is only a problem with very sensitive speakers. Nevertheless, I believe quiet background (high S/N ratio) gives better dynamics.

To remove the residual ripple, I had to remove a couple of volts from each power supply leg, and I have a lot of mosfet lying around, I used them. The only warning that needs to be made is that the zener stack was not drop too much power. If you look at NP's preamp (balanced zen) power supply, you'll see he used a 1.5K ohm resistor to bias the zeners. This resistor should be adjusted to run 10 to 20 mA of current for the stack (I use 3 zeners), while keeping in mind the total voltage. For example, if you use a 2K resistor is used for a 20V supply, the zeners will be biased with 10 mA (20/2K). The power dropped over them will be 10 mA x 20V, or 200 mW. If 500 mW zeners are used this is OK. However, the same resistor used with a 50V supply will bias at 25 mA and generate 1.25 W: too much for 500 mW zeners. A solution is to use larger zeners or increase the resistor to 5K. A reasonable solution is to make the bias resistor = volts/10, kOhms (5K = 50/10, kOhms).

Koy-

I'm not sure what to say about your simulation, but I get a lot lower THD in my simulation and in actuality. In fact, just this morning I powered up one new board with all my mods and it produces 20W class A with a THD of 0.0085% at 1W into 8 ohms at 1kHz (0.02% into 4 ohms). I presently have 2 pair of IRF240 mosfets running at 0.6 A each with +/- 25 V rails. Dissipation is thus 0.6 x 2 x 50, or 60 watts. Freq response is 10 Hz to 95 kHz (it was still flat at 10 Hz: my HP distortion analyzer only goes to 10 Hz; -3 dB at 95 kHz). Residual noise was 150 microvolts (input shorted); gain 20 dB; D/C offset 11 mV (input shorted).

Regards,

Robert
 

Nelson Pass

The one and only
Paid Member
2001-03-29 12:38 am
Actively loading the input diff pair gives a lot more
open loop gain, and the added feedback is very good at
reducing the distortion. I played with this many years
ago, but I didn't care for the sound as much. For anyone
prefering lower distortion, this works fine, but you
need to watch the frequency stability and you may find
yourself adding or adjusting frequency compensation.
 
Mr. Pass-

If I understand you correctly, I did experience oscillations occuring at about 500 kHz. These were cured by placing 0.01 microfarad capacitors between the gate and the source of the constant current IRF240 output devices (positive rail). This seemed rather large but I progressively increased the value until the oscillations stopped. Despite my misgivings that frequency response would suffer, the -3dB point was still around 95 kHz. For whatever reason (the layout is symmetrical) the signal (minus rail output) mosfets required no capacitors. Each mosfet, of course, has a 221 ohm resistor immediately adjacent to the gate.

By the way, after warming up for several hours, the residual noise was not 150 microvolts, but around 80 microvolts, getting near the limits of my distortion analyzer. The THD remained the same, around 0.0085% at 1W at 1 kHz.

As for the sound, I must admit there is a certain magic about the Zen amp (with no input devices) that seems missing from the Aleph amps. However, the Aleph amps can better control a difficult load without regard to the preamp peculiarities. The Aleph amps, with the current mirror control on the differential inputs mosfets, seems to provide an effortless sound. This amplifier, with regulated supplies and lower distortion, has an openness and smoothness about it that is so different from any amplifier I've used, and I've been fortunate to hear many such as Accuphase, Ayre, BAT, Boulder, Bryston, Burmester, Classe, Krell and Levinson to name a few. I prefer your amps.

On another note, I wish to thank-you for taking the time to answer our questions, but more to the point, thank-you Mr. Pass for sharing your designs with us. Such kindness and generosity is frequently missing in high end audio.

God bless,

Robert
 
Grey:

No I didn't try playing with C8. I'll give that a try. But with what I have to add, I 'm not sure it will help very much.


More ruminations:

On other matters, I like to describe how my final circuit performed, because I got a surprise tonight and I think I received some enlightenment on the functioning of Aleph amps in general.

First let me say that one of my goals in modifying the Aleph design was to more easily deliver power to lower impendances. Specifically, I wished to drive 4 ohm loads with more, rather than less power, than at 8 ohms. To accomplish this, I simulated an Aleph circuit after studying it (already having built a Zen amp).

I soon realized that, in addition to Q5 setting the bias on the CCS mosfets, Q1 and Q5 also behave as IV limiters. I therefore, attempted to optimize the resistors that affect the IV control of Q1/Q5 (all references in this post will be to Aleph 4 schematics).

Specifically, I changed R33 from 1K to 4.5K, R34 to from 523 to 1K, R20 from 221 to 4.5K, and R21 from 75 to 1K, and also R30 from 4.75K to 1.5K and R31 from 121K to 221K (these latter two were due to lower supply rails: I was using a 25V rather than a 48V supply). These resistor changes allowed better current delivery into 4 ohm loads as they relax the current limiting. At 1 kHz, with 1 ohm resistors for the output mosfets sources, the amp with produce 30 watts into 8 ohms and 55 watts into 4 ohms. The only problem is that the whole amp only works well like this at 1 kHz. At greater frequencies (> 7 kHz) it is only linear below 5 watts.

The problem seems to be that the amplifier starts to show class A-B distortion with greater power levels at higher frequencies. As I investigated, using both a real amp module and simulation, I found the following: R31-C6-R34 do indeed adjust the bias current to the CCS mosfet (Q6-8), but they do more than that. Their function is to adjust the bias using AC feedback (C6 blocks any DC) from the output to bias the gate.

If you take an oscilloscope probe (real or simulated) and monitor the signal at the gate of the CCS mosfet (Q6-8), you will find a signal as large as the output being feed into the so called "constant current source". What we really have with this design section is a modified push-pull. That is, the CCS is perhaps truly constant at low levels, but begins to behave as a active device at greater power levels, modulating WITH the signal. It is no longer a simple CCS; it is push-pull. And were we all not trying to avoid push-pull?

Therefore, the distortion I saw when driving the amp hard was typical class AB crossover distortion. Please correct me if I'm wrong, but I now believe that the only reason an Aleph is more efficient in class A than a Zen or many other class A amps, is due to the feedback through R31-C8-R34 shifting the amp from class A to class AB operation.

I tried to correct these non-linearities at higher frequencies by using a cap from the + to - rails, but this did not help very much. The only method that seemed to truly remove the class AB effect was to remove this feedback section (R34-C6-R31) and re-adjust the bias resistors. Of course, efficiency than drops to a typical class A at 25% or so. (If you remove the connection of R34 in simulation you will see trunction of the upper output signal peak; I found this in reality as well as simulation. This is another piece of evidence that the CCS is really behaving as a push-pull device.)

If I'm way off base, please clarify. Perhaps I'm too dense to understand. If there is any other way to efficiently drive low loads, help me out.

Robert
 
Robert

I hope I'm not stating the obvious or repeating something posted elsewhere, but your findings regarding the 'constant' current source are completely correct. The Aleph is a practical implementation of NP's patent for an 'Amplifier having an active current source'(American patent No 5,710,522). The method of circuit operation is detailed in the patent. The current source is modulated by the load current and gives an efficiency for the amplifier similar to that for a standard push-pull Class-A output stage.

By the way, you say that your component numbering is for the Aleph 4. I'm afraid that it does not concur with the circuit numbering on the diagram I have just downloaded from the Pass site (i.e. the one in the the Aleph 4 service manual). It always helps if we can refer to the same diagram :)

Geoff
 
Thank you for email. I completed my knowledge about current mirror. As far as I am able to understand it could work as local feedback, keeping current ( voltace across input MOSFEts and DC offset) at the same level.
Do your results come only from simulation? Did you notice any differrencies in sound? I made some simulations with Aleph 3. This amp appeared a little slow ( SR round 15V/us). What is experience?
Did you think how the current source (by signal end MOSFEt) affected the sound? best regards koy
 
Geoff-

I'm afraid I didn't look at the patent pages (I don't understand them anyway). As for the Aleph 4 nomenclature, I wrote my comment from another computer that has a different set of downloads. I looked over the ones on this other computer and you are correct, it doesn't match what I wrote. In any event, the R34-C6-R31 are the resistor-capacitor-resistor that is the feedback from the output to the upper current source. Using a copy (rev 0, 1996) of the Aleph 5 schematic, these are R19-C10-R21 for transitor Q5. The R30 I referred to is R17.

blmn-

I only used N-channel devices (IRFP240) in both sides. There are no P-channels in the output stage of the Alephs.


Koy-

I think the sound is more dynamic and how can I say it otherwise: smoother sounding. If the slew rate came down to 15V/us that still is plenty fast. (NP himself is the one who, if I'm not mistaken, calculated that music only slews at about 1-2V/us.)

As for the current source affecting the sound. Well, all I was saying is that the upper CS _measureably_ affects the sound by effectively turning the amp into a class AB at greater drive levels, especially at greater frequencies (most noticeable above 6kHz with my measurements). As for sonically being noticeable, I can't say. In the past, we did notice a definite modulation of the source upon the output with the Zen amp. This is because the feedback is not isolated from the input and so some preamps will interact with the Zen: driving the input and also modulating the feedback and re-driving, if you will, the amp. By having an input stage (Aleph), the preamp is better isolated from adversely interacting with the output stage. (As an aside, this should not be an issue with SOZ, as there is not feedback loop.)

Robert
 

blmn

Member
2001-02-01 2:43 am
.
Rljones,

Sorry, I made a mistake because I read you made a symmetrical "design" (not layout as you really wrote). I don't know this design very well. I remember now it was the same that I made an analysis in another thread. Since you have different functions for the devices (current source and modulator), I think it would be a good starting point to understand why you needed to add capacitance just for one gate.