Feb 27, 2007: Production σ11 boards are now available.
Feb 22, 2007: The official σ11 website is now up.
Feb 01, 2007: σ11, a single-rail version of the σ22, was described in Post #129, latest update is in Post #174.
Dec 13, 2006: Production σ22 boards are now available.
Nov 28, 2006: The official σ22 website is now up.
Nov 22, 2006: The development phase of the σ22 PSU is now complete.
Nov 05, 2006: The current schematic and PCB layout is in Post #55.
Oct 08, 2006: See Post #101 in the β22 thread for an illustration about heatsink options.

Introduction

The σ22 ("sigma 22") is a companion power supply for the β22 amplifier I recently announced. It is designed to be flexible and can output ±30V, ±24V, ±15V, ±12V or other voltages by changing the value of a single resistor (and in some cases, a zener diode). In conjunction with high output current capability, the σ22 could also be used with many other amps and circuits requiring a high quality dual-rail PSU.

The σ11 is a single-rail version of the σ22, suitable for use with amplifiers and circuits that require a high-performance single supply. Since there is only one rail, there is no tracking feature, but it is otherwise identical to the σ22.

As with the β22 project, I invite your participation. Perhaps your feedback would help improve the design.

The σ22's all-discrete design and complementary topology, along with MOSFET output devices thematically mirrors the β22 amplifier.

I decided against a two-stage regulator design (e.g., Jung super regulator and its variants, Gilmore Dynahi PSU, etc.), because that would not only increase complexity, but voltage drops across each regulator dictate a higher transformer secondary voltage, which increase total heat dissipation and reduce efficiency. Using only one regulator stage and making it perform well seems to be a more cost effective way to go.

Design goals

My design goals for this power supply is as follows:

1. A power supply that is worthy of the β22 amplifier, even though the β22's high PSRR makes it less demanding of PSU performance. To this end, I allow a "deluxe" approach to design but strive not to make things overly complex. Objective goals include low noise, high ripple/noise rejection, good regulation, and high output current capability.

2. Ease of parts sourcing: Choosing parts that are readily available where possible.

3. Avoid ultra-expensive ICs to help contain cost.

4. Can use either dual-secondary or center-tapped secondary power transformers.

5. Ease of construction: Given its nature this would not be a beginner's project, but the design will still strive to make the mechanical assembly and heatsinking/casing as uncomplicated as possible.

My original goal did not include the voltage tracking feature. However, as it turns out, adding this feature did not involve an increase in complexity. In fact, it allowed a reduction in parts count (if you look at the schematic diagram, you'll see that a non-tracking design would involve mirroring the positive supply side with an identical negative complement. However, by adding the tracking feature I could omit the voltage reference on the negative side, which is comprised of a CRD, zener, resistor and filter capacitor). So, this is a win-win situation and the tracking feature would make the σ22 also ideal for powering amps with poor PSRR.

Current Development Status

Development/simulation/prototyping are now all complete. The power supply is now "released".

PCB Availability

Boards are now available from AMB.

Schematic diagram

Note: The following schematic is out of date. See the News section above for a link to the current schematic, or visit the σ22 and σ11 websites.

Circuit description

AC input from the power transformer secondary gets full-wave rectified by discrete fast-recovery diodes D1-D4, each diode is bypassed with a snubber capacitor. The resultant positive and negative DC rails then feeds their respective voltage regulator sections.

The positive regulator consists of a discretely-implemented equivalent of high current output opamp. Q1 and Q2 form a classic differential long-tailed pair (LTP), loaded by a current mirror (Q5, Q6) to increase the open-loop gain. This is desirable in a PSU application because the increased global feedback would help keep the output impedance low and enhance overall stability.

Q9 is the VAS stage, loaded by a current source CR3, and compensated by capacitor C10. Two 18A power MOSFETs are connected in parallel to serve as the output "pass" transistors. The paralleling of two devices divides the heat dissipation, and to ensure that they would not fail under severe overcurrent conditions. This eliminates the need for current-limiting circuitry. Only a suitably-sized AC mains fuse is needed for protection against damage to the power transformer and other components. The reasonable maximum current capacity of this power supply is then basically limited only by the transformer, rectifiers, and the amount of heatsinking of the MOSFETs (and, of course, the AC mains fuse).

Edit: The voltage supply to the differential amplifier and VAS stages are now through a capacitance multiplier to provide a large boost in PSRR. See the current schematic diagram.

In the positive regulator, the "opamp" operates as a noninverting amplifier. The input of this amp is fed by a DC reference voltage provided by zener diode D5. The zener current is sourced from the CRD CR1 for high PSRR, and an RC filter formed by R1 and C9 (with a corner frequency of 1.6Hz) effectively removes any zener noise. The "opamp" then amplifies the reference voltage to the desired output voltage. The output voltage is determined by D5's zener voltage and the gain setting of the opamp (and can be varied by changing the value of R8).

The negative regulator has the same topology as the positive regulator, except that all transistors are complements of the positive version, and it operates as an inverting amplifier with a gain of 1. The input of this amplifier comes from the output of the positive regulator, thus making the output here an inverted version of the positive output. Whatever appears at the positive output is mirrored at the negative output, except it's out of phase. This is how the negative regulator "tracks" the positive.

The schematic shows both the input and output sides of the regulator having large reservoir capacitors, each bypassed by a film capacitor. There are two camps of thought with respect to the output side, some prefer high capacitance and others prefer less. The σ22 can work either way. In the low-capacitance scenario, σ22 has plenty of bandwidth to be "fast". The pre-regulator capacitance should be high to keep the ripple as low as possible.

Four sets of DC output terminal blocks are provided so that a single σ22 could supply a balanced β22 headphone amplifier (four amp boards). Two ground terminals are provided as the headphone jack ground return point if the amp is to be configured as a 2-channel passive amp (see my post on this subject for details). Addtionally, a terminal block is provided for an LED "power on" indicator light. The LED provides a small nominal load on the regulators (from the positive rail to the negative rail) even when there is nothing connected to the output. It will also help slowly discharge the rail capacitors after the power is turned off.

Heatsinking the MOSFETs

As with the β22, onboard heatsinks can be used for the MOSFETs. However, the PCB layout will also allow an off-board large heatsink to be bolted on for very high-current applications.

As mentioned above, the amount of heatsinking is an important factor in how much current this PSU is capable of delivering. For the β22 configured as a headphone amplifier, one σ22 PSU board can supply 2-channel passive ground, 3-channel active ground, or 4-channel balanced configurations. With a 4-channel β22 running 160mA quiescent current through the MOSFETs (for a total current draw of around 800mA), and if the voltage drop across the voltage regulator is 10V per rail, then the σ22's MOSFETs would dissipate about 4W each.

For a β22 operating as a speaker amplifier, if you want to use the σ22 then there should be one σ22 board per channel, and an off-board large heatsink is recommended. You may also use individual σ22 boards per channel in a headphone amp, but that is overkill and not necessary.

[Edited by amb on 11-11-2007 at 03:12 PM.]

---

[Quoting masantos]

Amb, I've been reading a lot about gainclones since I might build one in the near future. I have read many threads that say that the MUR860 performs better than the 820 one. It is the same diode but with higher voltage rating, which seems to improve speed and reduce noise. Maybe I would suggest you change the default part to the MUR860.

---

At any rate, I doubt that there would be a real tangible difference between the two in this application, and the MUR860 is a little more expensive.

Once I reworked the rectifying section of my Dynahi PSU that I bought almost completely assembled. I removed the high voltage ultra fast ("snapping") rectifyers and put soft recovery hexfreds in, additionally I changed the snubbers from small ceramic caps to large 0.47uF film caps. That change was audible, not dramatic and rather subtle but audible. Nevertheless I wouldn't go overboard with extreme spec hunting. I think the MUR820 (used them in quite some of my PSUs) are a decent and affordable choice, just perfect for the default recommendation.

As for the snubbers I'd like to use film caps of about 100nF to 330nF instead of 220pf, combo pads for 5mm and 7.5mm lead spacing would be very appreciated.

What do you think about adding a RC (e.g. 0R5 with 100nF) in parallel to the 4k7 and 1uF caps at the output to flatten the hf impedance of the PSU, particulary with current surges ?

Really nice PSU, unfortunately I can only comment on part selection and can't be helpful with the design. I'm looking forward to building it.

---

[Quoting aos]

By the way amb, what's the bandwidth of the regulator (what's the frequency when the PSSR starts to roll off)?

---

Edit: After doing some AC sweep analysis I decided to eliminate the C12 and C13 feedback capacitors as shown in the v0.1 schematic above. In the positive regulator, this capacitor's effect is marginal because it cannot reduce the closed loop gain below unity. Instead, I'll rely on the C10 and C11 compensation caps to set the bandwidth. Aos is right in that the compensation caps' value will have to be high in order to maintain stability when large reservoir caps are used at the output. This change will be reflected in the next σ22 schematic revision. Here, the discrete implementation has an advantage over IC regulators, in that one could "tune" the bandwidth by varying the compensation capacitors' value. When very low rail capacitance is used at the output, the regulator can be tweaked to have very wide bandwidth.

[Edited by amb on 09-15-2006 at 04:12 AM.]

For the β22 I suggest staying with the default values as shown.

1. Added capacitance multipliers to supply the reference zener, differential and voltage amplifier stages. This gives a large boost in PSRR for a small increase in parts.

2. Re-introduced the compensation capacitors in parallel with the feedback resistors. These still provide useful response shaping without resorting to overly large miller capacitors at the voltage amplifier stage.

3. Eliminated the large output rail capacitors. It is now assumed that the amp board will have such caps. I now recommend using moderate output rail capacitances to allow the regulator to operate with wider bandwidth (See the notes in the schematic about choosing compensation capacitor values based on the output rail capacitance. Interpolate for in-between values).

If this were a class AB amplifier design, I suppose it would be a contest between the impedance of low-ESR amplifier rail capacitors and the output impedance of the regulator. In other words, when a quick, loud bass note comes along, which source would deliver current to the amplifier the fastest? The capacitors would probably win.

But since the β22 is designed to biased deep into class A even into very low-impedance loads, there should not be any variance in current demand even at signal peaks, so no need at all for reservoir capacitance. So therefore the regulator, in this application, is strictly being used to manage power-supply noise and fluctuations -- i.e. it's regulating out problems at the supply side, not at the power-demand side.

Therefore, if I'm looking at this properly, it seems to me that there's almost no need for much amplifier rail capacitors on the β22, and it's worthwhile to crank the regulator up to its maximum bandwidth.

Agreed?

could you explain/develop the issue of regulator bandwidth?

if you feel this shouldn't be discussed here I can start a new thread, but I would like to understand this subject more.

Thanks in advance,

Manuel

Revision 0.3 is much improved with the capacitance multipliers cleaning up the current for the reference zener and the first stage of the amplifiers. Prior to this enhancement, I wanted to comment that, IMO, the supply rejection was inadequate on the zener current source.

Have you considered making the amplifer a cascoded single stage only? The bandwidth would increase substantially with no loss in gain. It would go from a three pole system to two. Swing and heat dissapation issues are possible limitaitons for this though, depending on how it would be biased.

---

[Quoting masantos]

could you explain/develop the issue of regulator bandwidth?

---

As I eluded to before, there are two schools of thought about load regulation. One such is to put a lot of rail capacitance at the output of the regulator, such that the stored charge would supply any instantaneous current demand the amplifier might need. This works well, and the voltage regulator has a comparatively "easy" job of simply keeping the rail capacitors charged. As such, the regulator does not need very wide bandwidth. Indeed, since the regulator is just an amplifier, and amplifiers tend to become unstable with a highly capacitive load without heavy compensation (which reduces the bandwidth), a large rail capacitance at the regulator's output requires a limited regulator bandwidth. Almost all IC regulator chips are internally heavily compensated to handle large rail capacitors, and thus have limited bandwidth.

The other school of thought is to use minimal rail capacitance at the regulator's output. This makes the regulator responsible for supplying the fast current demands from the amp, and therefore the regulator must have wide bandwidth in order to keep up and not to starve the amp. The use of small rail capacitance allows the regulator to be much less heavily compensated to achieve stability, which, of course, is needed for the wide bandwidth. If a wide bandwidth regulator is used with too much rail capacitance, it would most likely oscillate.

For amps such as the β22, there are a few hundred µF of rail capacitance onboard. This is not very high, but not ultra-low either. I feel that it's a good idea because the power supply is not on the same pcb as the amp and the PSU may live in a separate case with a umbilical cable, which can have a small amount of impedance. The local capacitance will help dampen and reduce such impedance and improve the amp's transient response. Without it the amp may become unstable. For σ22 v0.3 I deleted the large reservoir caps at the regulator output, so aside from small bypass caps, the capacitors in the amp is the only post-regulator rail capacitance. Hence, my approach now is sort of a "middle of the road" one, where the output rails have moderate capacitance yet the PSU isn't too overly compensated for decent bandwidth.

Since the β22 amp operates in pure class A with virtually any headphone load, its current draw from the PSU is constant. Thus, either solution should yield excellent performance, but the wide-bandwidth way eliminates a couple of large capacitors, which saves cost and pcb real-estate.

Also, since the σ22 PSU is fully discrete, you can change the amount of compensation to set the desired bandwidth. The notes in the v0.3 schematic diagram provides a guideline, which was derived from simulation. When I have a real circuit built I will test it and refine the guideline if necessary.

Hope this is good enough explanation. Others please feel free to chime in with more.

---

[Quoting snoopy]

Revision 0.3 is much improved with the capacitance multipliers cleaning up the current for the reference zener and the first stage of the amplifiers. Prior to this enhancement, I wanted to comment that, IMO, the supply rejection was inadequate on the zener current source.

---

---

[Quote]

Have you considered making the amplifer a cascoded single stage only?

---

[Edited by amb on 10-01-2006 at 12:48 AM.]

(if this impedance plus cable impedance is lower than the impedance of a bank of low-ESR reservoir capacitors, then it's maybe an argument in favour of the low-rail-capacitance approach...)

---

[Quoting dougigs]

AMB, did your simulations tell you the output impedance of the regulator at audio frequencies?

(if this impedance plus cable impedance is lower than the impedance of a bank of low-ESR reservoir capacitors, then it's maybe an argument in favour of the low-rail-capacitance approach...)

---

I too tend to view the regulator as a buffered voltage reference. This seems to be a productive way to go about designing one, at any rate.

Wading into the bandwidth debate, I think if you are going to the trouble of designing a high performance regulator, its bandwidth should extend at least over above the audio band, and preferably to just above the bandwidth of the amplifier it serves. To get the most benefit for the low impedance and avoid oscillation problems, it should also be built as part of the amplifier circuit, i.e. on the same PCB and as physically close as possible to the amplifier. Rail capacitance should be kept to an absolute minimum.

In terms of ripple/noise rejection, the noise can get into the output through the voltage reference or the error amplifier since both of these connect to the input voltage, as well as through the pass element. The design goal is to get both the error amp and the voltage ref noise just low enough so they dont show in the output above what would get through the pass element anyway... but not any lower. Overkill needlessly increases the impedances feeding these components, which can lead to more circuit noise.

Sort of related to that, can you demonstrate that your discrete circuit is better than a decent low noise opamp in this application? That's not an accusation, I'm just curious as to how they compare. For the record I really like the simplicity and ease of configuration of your discrete excecution.

Lastly, the huge current capacity of the MOSFETs seems to me a bit of a waste. Large MOSFETs have big gate capacitances that may be difficult for your circuit to drive, and when the output current isnt going to be much more than 100 mA or so a more proportionate part seems to be called for. Heat isnt going to be an issue, so why not a single low current MOSFET? Output protection isnt really needed if the regs are on the same board as the amp anyway, only if you have a umbilical can it be dangerous.

Incorporating the regulator into the amp was something I had considered, but I decided against it for the following reasons:

1. For space reasons - the main amp PCB is already crammed full of parts and I want to keep the board size down in order to fit it in some of the off-the-shelf cases, such as the Hammond 1455T220x.
2. For flexibility reasons - even though this PSU was designed to be a companion for the β22 ampifier, I want it to be usable with other amps as well.
3. For physical isolation - Having the entire PSU in a separate case than the amp, in particular the power transformer and rectifier diodes, minimizes noise intrusion into the amp. It also allows a more optimal grounding arrangement (PSU case to AC earth, amp case to signal ground) without the need for hacky ground loop breakers.

The use of discrete elements in this PSU was originally for thematic similarity to the amp, but as it turns out there are good reasons to go this route. There are plenty of low noise IC opamps out there, but few with high enough supply voltage handling capability to work in this circuit. There are high voltage opamps but they tend either to be less stellar in noise performance, difficult to buy everywhere, or are very expensive. Further, implementing this with discrete parts gives the possibility to set the bandwidth by changing some compensation capacitors, something not always easily done with IC opamps. In short, the all-discrete solution neatly lets us have our cake and eat it too.

As for output current capability, this circuit can supply four β22 amp boards, which, when biased to 200mA each, would draw a bit over 800mA total at quiescent. If used as a speaker amp, the dynamic current could exceed that by several times.

Using two paralleled MOSFETs per rail, as mentioned in the initial post, was not so much for increasing the current capability, but to divide the heat dissipation. If the pre-regulator voltage is 10V higher than the output voltage, then the pass MOSFET would have to dissipate 8W at 800mA draw. 10V headroom is actually not too much, because the error amplifier is powered from the pre-regulated rail via a capacitance multiplier, which drops a little voltage, and since the pass MOSFET works as a source follower, it cannot output any higher than the maximum swing of the error amplifier minus the MOSFET's Vgs (which is about 4V). Let's say the cap multiplier drops a volt and the error amplifier can swing pretty close to its rail, then we have less than 5V left for line regulation headroom.

8W is a lot for a single TO-220 device to dissipate continuously and would require a large heatsink. We might as well let two MOSFETs share the load so that each only bears half of that dissipation, making the onboard heatsink solution feasible. The MOSFETs are relatively inexpensive so IMHO is not a "waste".

The board is 6" W x 3.5" L. It is exactly twice as wide and half as long as the β22 amplifier PCB. The reason for this is I am also aiming to fit this board into a Hammond 1455T220x case, but the board must be cut this way in order to leave enough room for a power transformer. C5 and C6 are now each two 2200µF capacitors in parallel. This is because the Nichicon UPW/UHE and Panasonic FC 50V lines do not go up to 4000µF+.

The heatsink arrangement is identical to the β22, with the same hole spacings, etc.

There is a ground plane on each side of the board (not shown) to keep ground impedance to a minimum.

The schematic diagram:
Fixed a mistake in v0.3 (the top of R11 should not be connected to ground), and other cosmetic adjustments.

The PCB layout:

[Edited by amb on 10-08-2006 at 01:16 AM.]

Changes:
- Moved the output terminal blocks inward and re-routed the output traces to the outside. This prevents the output traces, which are on both layers, from forming a barrier around the ground plane around the ground connections at the terminal blocks.
- Widened most high current traces.
- Done a net-by-net check and fixed a few errors.
- Other minor adjustments to position of parts and traces.

[Edited by amb on 10-09-2006 at 06:23 AM.]

Sorry, is me again... still some suggests from my mind.

1.C5 &C6 will split to A&B? possible add more config for one dia.30~35mm big boy?

2.May CR2 more close to center, and C11 & C12(470uF) can have more space put mmm... fat caps. (C11 can turn 180 degrees.)

That will be sweat~

Thanks!!!

Pal