Introduction

I had previously posted the idea about creating a high-power version of the β22 as a dedicated speaker power amplifier. It is quite a logical step to take, because β22 is already an excellent low-power speaker amp and has most of the ingredients of a sophisticated high power amplifier. The only limitations to more power are the power supply, rail voltage, output device capability and power dissipation concerns.

Thus, I set a goal of 100W into 8Ω and went to work. The result is shown here, and it's called the β24. Currently, I consider this a preliminary "work in progress", even though a lot of work has already been done and PSPICE simulation results look excellent. Your feedback and comments, as usual, are very welcome.

Overview

β24 is derived from β22, and it should be immediately evident upon seeing the schematic. The input complementary differential stage, the second complementary differential VAS (voltage amplification stage), and output stage are all essentially identical to the β22.

There are two ways to go about increasing the power output in a linear amplifier.

The first method is to raise the supply voltage to increase the available voltage swing. This often requires most of the active devices to be replaced with higher-voltage variants.

The second method is to employ balanced differential output (also known as "BTL" -- bridge tied load) which doubles the voltage swing, and theoretically increases the power four-fold without raising the supply voltage. However in most instances this requires two amplifiers per channel, doubling the parts count.

Both methods require appropriate upgrades to the output stage, heatsinking, power supply, and various other tweaks.

Due to the differential VAS in the β22 topology, we already have both signal phases available. In the β22, the inverted phase is not used. The differential design was employed mostly for its distortion-canceling properties. By adding just another output stage to the circuit, fully-differential outputs can be realized without doubling the number of parts. Thus, this was the primary method chosen for the β24. The supply voltage to the input and VAS stages are also raised slightly. In addition to more power, balanced output drive has other benefits that I will mention below.

Using the differential output design also allows the use of most of the same type of active devices found in the β22. The output devices have to change, obviosuly, because β22's MOSFETs do not have sufficient power dissipation (and low enough thermal resistance) to be used in the β24. I have investigated and ran SPICE simulations using a number of options, including Darlington and Sziklai BJT pairs (using various devices), but ultimately came back to MOSFETs due to their lack of secondary breakdown concerns, and the negative temperature coefficient, which are important for reliability. I also contemplated the use of many paralleled pairs of smaller MOSFETs versus one single set of large MOSFETs, and decided on the latter. It would make PCB layout and mechanical construction a lot easier.

Since we have a differential input stage, it is not difficult to rework the amplifier (along with the two output stages) into a versatile fully-differential amplifier. Having a floating differential input allows the amp to be easily configured as having balanced or unbalanced input, without the need for additional phase inverting circuitry. It could even be switchable between balanced or unbalanced input.

The supply rail capacitance multipliers found on the β22 were eliminated in the β24. Instead, I chose to have two separate sets of supply rails. The input stage and VAS are powered by the quiet and highly-regulated σ22 PSU, while the output stage is powered by an unregulated supply consisting of eight reservior capacitors. This decouples the high-current output stage from the low-signal stages. It also removes the need for a large, high-power voltage regulator for the output stages, which simplifies the power supply requirements, not to mention a substantial reduction in heatsinking/size/cost. The σ22 only needs to supply about 100mA per channel, which makes its heat dissipation very manageable.

The differential nature of the amplifier means that any ripples on the unregulated output stage supply rails are canceled. The output stage also has extraordinarily high PSRR, under simulation in non-differential output condition, 1V ripples at 120Hz on a rail resulted in less than 80nV at the output, which would be lower than the amp's noise floor. Under differential output conditions, no ripple component was seen at the output at all.

Schematic Diagram

Here is the β24 schematic diagram. Most parts IDs have been reassigned with new numbers from those in the β22 due to the additional output stage and other changes.

(Note: the schematic shown here has been changed. See the "News" section above for a link to the most up-to-date version.)

The main amplifier circuit is shown in the diagram above, the differential inputs are denoted as I+ and I-, while the differential outputs are O+ and O-. The lower left corner inset shows the twin, cross-coupled feedback loops and the output zobel network. For balanced input, the Input+ and Input- connections should be connected to pins 2 and 3 of the XLR input jack, respectively, while pin 1 should be connected to chassis ground. For unbalanced input, Input- should be tied to ground, and Input+ used for the signal. A switchable balanced/unbalanced input scheme is shown in the diagram below:

Input Stage and Voltage Amplification Stage

The input stage and VAS are virtually identical to that in the β22, having the same dynamically cascoded topology and class A operation. The VAS as drawn in this schematic is shown with its left and right sides split apart, surrounding the input stage, but upon closer examination you'll see that there is no real change.

For a description of these stages, please see the β22 announcement post as well as the Official β22 website (under "Tech highlights" and "Schematic").

The only change from the β22 design in these stages are as follows:

- The global negative feedback scheme is changed to allow twin feedback paths from the differential outputs, and to provide configurable balanced/unbalanced inputs. With this change, you could think of the β24 as an all-discrete, high-powered implementation of the OPA1632 or THAT1606.

- The power supply rail voltage for these stages have been increased from ±30V to ±36V. This increases the voltage swing headroom and ensures that when clipping occurs, it happens at the output stage rather than the input/VAS stages.

- Due to the increased supply voltage, and also due to the obsolescence of the J503, the CR1-CR4 current regulating diodes should all be 1N5291 or E-501, which are rated at 100V.

- The VAS quiescent current is reduced from 15mA to 9mA. This is because the VAS no longer drives the output MOSFETs directly, and does not need the higher current for gate charge purposes. The lower current reduces the power dissipation of the transistors in this stage, which is necessary due to the slightly higher rail voltages.
(EDIT: This has been changed. The VAS now does drive the output MOSFETs directly like in the β22, and the VAS current has been increased by a change in initial setup bias adjustment target rather than changing any component values.)

- The open loop gain of the amplifier is increased from about 58dB in the β22 to 64dB in the β24. This is in proportion to the increase of the closed loop gain from 8 to 20, which is a 8dB difference. 64dB is still a relatively low open loop gain (compared to 120dB+ in some opamps), which allows a moderate amount of global negative feedback.

- The compensation capacitors C3-C6 values have been re-tuned for the new output devices and gain settings for optimum bandwidth/speed and stability.

- The DC offset adjustment trimpot is moved to between the N-channel input JFETs' source pins. This allow the differential DC offset to be adjusted. No common-mode DC adjustment is provided because such a DC offset is not "visible" by the load.
(EDIT: A CMFB circuit has been added to automatically control common-mode DC offset.)

Vbe Multiplier

The complementary Vbe multipliers Q17/Q18 and Q19/Q20 provide biasing for the output stage, and are unchanged from the β22. The resistor and trimpot values have been modified to accommodate the new output stage design.

Output Stage

The output stage retain the dynamically cascoded design from the β22, utilizing two pairs of enhanced-mode power MOSFETs. The output devices have been changed from the β22's IRFZ24N/IRF9Z34N to IRFP140N/IRFP9140N. These are larger TO-247AC devices, rated at 100V, 33A/-23A and 140W, used in the Q29-Q36 positions. The cascoding divides the power dissipation seen by each device, so we will be operating them well within their safety margins, as long as sufficient heatsinks are used.

To handle the higher gate capacitance of the larger MOSFETs, a BD139/BD140 complementary emitter follower pre-driver stage (Q25/Q26, Q27/Q28) was added, which provides a low-impedance, high-current (50mA) source to drive the MOSFET gates. These transistors, with their high Ft of 190MHz, provide a good combination of bandwidth/speed and power capability for the job, and are easy devices to source.
(EDIT: The pre-driver stage has been eliminated.)

In the β22, the current sources used to bias the cascode MOSFETs were implemented with PN4392 JFETs at 15mA. For the β24, a similar function is provided by Supertex depletion-mode MOSFETs DN2535N5 (TO-220 package) running at 35mA. Due to the increased current, the D7/D8 and D9/D10 zener diodes must be rated at 1W. BZX85C12 or 1N4742A could be used.
(EDIT: These zener diodes have been changed to 1N5931 18V 3W units)

The output resistors R45-R48 are specified to be Caddock MP915 or MP930 series, which are rated 15W and 30W, respectively. The actual power requirement is more like 5W, but not many other 0.2Ω options exist for non-wirewound types. The Caddocks are superior resistors requiring a relatively small footprint on the PCB.

Heatsinks

The output MOSFETs should all be mounted on substantial heatsinks, I recommend a thermal resistance of 0.5°C/W or lower per bank of four MOSFETs. Unlike the β22, no onboard heatsink option will be available for the output MOSFETs, because such heatsinks would not be adequate for this level of power.

Onboard TO-220 heatsinks with a thermal resistance of 25°C/W or lower should be used on the DN2535N5 depletion-mode MOSFETs Q21-Q24. Smaller (35°C/W or lower) onboard heatsinks should also be used on Q25-28.
(EDIT: Q25-Q28 have been eliminated.)

Power Supply

For best performance, I recommend independent power supplies per stereo channel. Of course, this would be de facto for a monoblock configuration.

The Vreg+/Vreg- rails on the schematic are to be supplied by a σ22 PSU, configured to ±36V. A 15VA, 35V+35V transformer per stereo channel should be adequate for this purpose. If a common transformer is to be shared between both channels, use a 30VA unit.

The high-current V+/V- rails are unregulated, with eight onboard rail-to-rail 2200µF reservoir capacitors. These rails should maintain at least ±30V under full load, so that the 100W output power into 8Ω rating is not compromised. I recommend a 30+30V, 250VA (or higher) power transformer for the unregulated rails per stereo channel.
(EDIT: In addition to the onboard capacitors, there needs to be at least 4700uF x2 per channel of filter capacitors after the bridge rectifier.)

Feedback / Voltage Gain

The voltage gain of the amplifier is set by the ratio of R2/R1 for the non-inverting side and R4/R3 on the inverting side. The values of these resistors should be closely matched for best performance. The resistor values shown provide a voltage gain of 20x (26dB), which is appropriate for a 100W-class amplifier. An input voltage 1.4Vrms will achieve 28.3Vrms of output, which is 100W into 8Ω.

These resistor values should not be altered, because the C1 and C2 compensation capacitor values are carefully tuned for these resistor values, and chosen for optimum square wave response and bandwidth for the the transistors used in the amplifier.

Due to the cross-coupled twin feedback scheme, the R1 and R3 resistors determine the input impedance of the amplifier, just like they do in an inverting opamp topology. As a consequence, we are constrained by a couple of factors in choosing the resistor values. If we want high input impedance, then R1 and R3 must have high resistance. But for a target gain of 20, R2 and R4 would then have to be 20 times higher than R1 and R3. Too high a resistance introduces noise, and stray capacitances become an issue. As a compromise, I chose 5KΩ (4.99KΩ) and 100KΩ. An input impedance of 5KΩ is relatively low, so a preamplifier with low output impedance (less than 500Ω) should be used. Most pro-grade balanced gear should be able to drive a 5KΩ load with ease. A "passive" preamp without an active output buffer, on the other hand, is not recommended.

Any good headphone amplifier with low output impedance should serve as an excellent preamp for the β24. A 2-channel β22, with its commensurate performance, would be a superb companion preamp. A preamp with fully-balanced outputs, with fully-balanced sources would of course be an ultimate match for this amplifier.

PCB Layout / Mechanical

See the News section above for links to the latest PCB layout work.

Board Availability

A limited number of v0.6 PCBs are available. PM me for details.

[Edited by amb on 06-29-2008 at 08:39 AM.]

Here is a 1KHz sine wave, driven to clipping. The output stage power supply is at ±30V, and the output clipped at 93Vpp (33Vrms), which corresponds to 136W into 8Ω.

Here is the closed loop frequency response and phase response. The -3dB point is at 1.6MHz, and the phase remains linear throughout the audio band, beginning to deviate slightly away from linearity above 100KHz.

The open loop gain is shown in the diagram below. Unlike opamps which typically show a rolloff beginning as low as 50Hz, the β24's rolloff -3dB point is 46KHz.

This is the output impedance of the amplifier. It remains below 0.005Ω throughout the audio band, increasing to 0.012Ω at 100KHz, 0.24Ω at 1MHz, and only 3.4Ω at 10MHz. This translates to an extraordinary damping factor of 1600 within the audio band. In reality this would largely be limited by the hookup wiring and speaker cables.

[Edited by amb on 01-21-2008 at 08:34 AM.]

This one is with the amplifier delivering 2.53Vpp (0.1W into 8Ω). The 3rd harmonic at 3KHz corresponds to 0.0008%. There are also spikes at 5th, 7th and 9th order, but all of these are so low as to be completely inconsequential.

At 8Vpp (1W into 8Ω), the 3rd harmonic is 0.002%. Again, smaller spikes are also seen at the 5th and 7th harmonics, but these are also too low to be of concern.

At 25.3Vpp (10W into 8Ω), the 3rd harmonic is 0.001%. The only significant harmonic is the 3rd.

Finally, at 80Vpp (100W full power into 8Ω), the 3rd harmonic is 0.025%. Again, the 3rd harmonic is the only dominant spike and is very low.

[Edited by amb on 01-21-2008 at 08:38 AM.]

Delivering 80Vpp (100W full power into 8Ω):

When examining an expanded X-axis of the square wave edge (not shown), the amp slewed from 10% to 90% of the 80V in 250nS, which corresponds to a slew rate of 290V/µS. This is very fast.

A word about these results. SPICE simulations always produce better results than reality, because all transistors of the same type are perfectly matched, and effects such as trace inductance and capacitive coupling are not taken into account, etc. Nevertheless, my experience in the past has shown that the transistor models I have are good, and these simulations provide a very good approximation of the actual performance.

[Edited by amb on 01-21-2008 at 10:05 AM.]

Regarding the board layout and size, it would be awesome to have everything cased inside a rack sized enclosure(300mx400mm, if I'm not mistaken), placing the boards on the side and the trafo in the middle( and any other circuitry). Using an encapsulated trafo would minimise noise.
Just someting to consider...

Will such an amp require a soft start circuit, since it uses such a large trafo? If you offered this circuitry in the same board it would allow a neater wiring with no extra boards required.

What about driving lower impedance speakers? Will it drive 4ohm or even 2 and 1 ohm subwoofers?

[Edited by masantos on 01-21-2008 at 12:27 PM.]

A couple of questions:

- Is it worth considering a gain option lower than 20? A lot of us are getting plenty of voltage gain from our line stages; I'm often tempted to think that power amps should be current-gain-only.. given the high output levels of most modern signal sources, and the fact that most high-quality line stages are providing gains of 8 to 12, I suspect that we don't need that much voltage gain in the first two stages. More realistically, would you consider trying a feedback RC combination optimized to a balanced gain of 10?

- What bias level and output Iq are you using in your simulations? And at that level, at what power level is it going AB? There will be some (not me) who will be tempted to build this with lower supply voltages and higher bias currents in order to get a bigger chunk of class A. Thoughts?

- I know you were holding out for matched-pair JFET packages in your original β22 design... are these complete vapourware as an input option now?

- Before you completely eliminate the capacitance multiplier on the input stage, there is a school of audio thought which holds that regulators followed by capacitance multipliers are, in and of themselves, a better-sounding combination. Maybe audio voodoo, but worth giving a listening test anyway.

- It's worth reminding people that, as appealing as "monoblock" sounds to hardware-heads, there are serious advantages in a design like this to having a much larger, single transformer shared between the two channels (each having its own rectifiers and capacitor bank) - - a 500VA transformer will give you a 4 or 5 per cent regulation level, compared to 10 per cent for 250VA. So a huge, shared tranny serves as a stiffer voltage regulator than a monster capacitor bank.

[Edited by dougigs on 01-21-2008 at 03:57 PM.]

You do have a fundamental issue that must be resolved however. The problem is that the common mode closed loop bias is not explicitly defined. This is because you are making a fully-differential amplifier and not using two single-ended amplifiers. When two single ended amplifiers are used, one of the input terminals of the amplifier is forced to a known DC potential for bias. The loop then tries to force the other input the same voltage. Thus the common-mode DC bias voltage is established. However, when going to a fully differential structure like you have done here, that extra input terminal forced to a known voltage is lost, and consequently the well defined bias points for both the inputs and outputs. In order to establish well defined closed-loop common-mode bias voltages, you must add a common-mode feedback (CMFB) loop. This is not optional for a fully-differential amplifier. Period. Moreover, the loop should be robust, well compensated and fast. The bennefits include the aforementioned stable bias point and suppression of even order harmonics.

One of the simplest ways to implement a CMFB loop is to resistively take the average of the two output terminals and, with a differential pair, compare it to the desired DC common-mode voltage. A control current would then be mirrored and fed back into the main amplifier, slightly adjusting the bias currents. Your design choices complicate this because you don't have tail current sources at the rails.

A couple of other points...
* What do you have against dB scale plots for FFTs and frequency response?
* Q21 & Q23 should be P-type to better match the impedances seen looking up and down for symmetry. You've gone through the trouble with the Vbe multipier, why not here.
* Otherwise, everything looks well thought out and should work as you expect.

Once again, nice work and thanks for sharing.

It explains very graphically why these things are necessary, and explains the calculation of CMFB points. Good catchup for us hobbyists.

[Edited by dougigs on 01-21-2008 at 04:42 PM.]

As for a soft-start circuit, it's not a bad idea but that's a power supply detail we'll visit later.

Lower impedance speakers are potentially a problem. Since the effective impedance "seen" by each output stage is half the real impedance, going much lower would quickly exceed the amp's current/power safety margins if you expect to play at a loud volume continuously. Of course in a home stereo setting this is usually not the case...

dougigs, Lower gain is possible, but consider that even with high line-out levels of around 2Vrms from CDPs, and the fact that most preamps have a gain of around 10 (but at a "normal" listening level the volume knob is set where there is usually quite a net loss of gain), we need sufficient gain in the power amp to achieve maximum power. As stated, a gain of 20 will get you 28.3Vrms (100W into 8 ohms) ar around 1.4Vrms input, and that's a fairly good ballpark.

Thanks for the link to the CMFB lecture. I also found some others to ponder on.

Snoopy, Thanks for your comments. Yes, the common mode bias issue had crossed my mind and I need to look into it further.

I don't have anything against dB plots, it's just that sometimes it's difficult to see relative magnitudes with a log Y axis, sort of like picking a tree from the forest.

Unfortunately, there is no complementary P-channel depletion mode device to the DN2535 (or similar) that I could find, so I had to use the N-channel device for the CCS both top and bottom.

[Edited by amb on 01-21-2008 at 06:59 PM.]

If necessary, I will ask Mr. Pass for permission to use this design. It'd be amusing if I had actually managed to (independently) come up with something that's been patended by the revered Nelson Pass.

I sincerely hope that Mr. Pass will be generous and let us DIY junkies persue our hobbies without undue hardship.

If you look at the basic schematic on his patent, if you collapsed R40 and the associated mess into a single current sink you'd be looking at basically every single solid state diff amp ever made.

TI considered it wasn't worth fighting the patent.

An older thread about what was exactly covered : http://www.diyau...6861&perpage=25

Nelson's always been very friendly with the DIY community in my somewhat limited experience, so that gives hope for permission if permission is needed. I'll be watching development closely

He says that the design does fall under his patent, but their policy is to not interfere with non-commercial efforts, and that I should feel free to publish the design and make circuits boards and such.

Many thanks Nelson, for being such a good sport.

Pretty good stuff in there. His current-source patent is a great read, as are his methods for class-A bias (dual Vbe multipliers).

You'll also find at least one patent held by amb in there....

Now back to our regularly scheduled amplifier project.

[Edited by dougigs on 01-22-2008 at 05:36 PM.]

---

[Quoting amb]

Nelson Pass has responded to my email, and it's a green light!

He says that the design does fall under his patent...

---

Am I crazy or am I just not seeing the folded cascode or what? You could claim the amp is 'supersymmetric' or w/e, but I don't see how it falls directly under the patent without folded cascodes.

---

[Quoting GridStop]

Am I crazy or am I just not seeing the folded cascode or what? You could claim the amp is 'supersymmetric' or w/e, but I don't see how it falls directly under the patent without folded cascodes.

---