813 Matrix Amplifier


Since I now had two successful Matrix amps, I decided to do a high power version, using transformer coupling and reasonably high quality parts. (I settled on Lundahl transformers for the audio iron). This design would use a single 813 per channel. What I achieved was 66 watts per channel without overdissipating anything. The idle point chosen is 800 volts at 100mA (80 watts dissipation per tube). In a normal SE design, this would limit the maximum power to something lower than 40 watts, but in Matrix, additional power is available as has been previously shown.

I also started out with the same 6MB8 tubes for the driver, and ultimately changed to 6GX7. The 6GX7 provided slightly improved definition and low level "niceness". This will be discussed later.

I also modularized all the power supplies. Every supply in this amplifier is regulated.

Since the Lundahl output transformers have multiple secondary taps, the Matrix manipulation is actually simplified, and it is possible to ground the low side of both outputs (unlike the tapped transformer, where the + of one side is ground).

With an input transformer, the transformer can do the phase inversion, so the "odd" stage is not needed.


This is a complex amp, and there are several schematics for the overall amp. These are .PDF files. The schematics are linked in the table below:

813 Matrix Amp Schematics
813mat.pdf Main Amplifier schematic
813matng.pdf Negative and Filament Regulated Supplies - the negative 100v regulator is "master"
813mathv.pdf High Voltage regulators (plate and screen)
813matvu.pdf The amp has a pair of UM81 "magic eye" indicators. This is their schematic.
813matlv.pdf Low voltage regulators.
813matic.pdf Interconnection and Front Panel (LEDs) connection of all the pages
813matin.pdf Alternative "loudness compensation" input circuitry

The amplifier is built in 3 modules:

These can then be individually built and tested (except the + supplies require the -100 volt output) and then installed in the chassis. There are LED monitors of all the major voltages produced. Each LED is "bypassed" by a 5 volt zener diode so that nothing dastardly happens if the LED fails, and the individual modules can be tested without the front panel LEDs hooked up.

The amp itself as built looks like this:

The front panel is a 19 inch "rack" panel 7 inches tall (4U). The sub-chassis on the left (with the 6080s and the light bulb) is the low and high voltage regulators. The aluminum sub-chassis in the middle is the negative regulators and filament supplies and the bias adjustments are obvious on the top. The chassis on the right with the 813s mounted horizontally is the amplifier module. The magic eye tubes are mounted to the front panel along with the LEDs, volume control and input connectors. (Yes, the amplifier is "on" in this picture. Note the UM81 softly glowing, the 8 LEDs, and the light bulb glowing as well as the tubes).

After having built it this way, I'd still recommend standing the 813s upright for heat dissipation purposes.

Circuit Description

The amplifier circuit is reasonably simple. Input transformer couples via level control to grid of the driver tube. Since Matrix requires one channel inverted, the input side of the right channel input transformer is connected opposite of the left channel, providing the inversion. The driver stage is cathode biased and not bypassed. This tends to increase the output impedance. Since the output stage has local feedback applied to the plate, this is desireable. The feedback summing node is the plate of the driver tube and driver transformer. The impedance of feedback summing node is generally low, which is what is desired for the transformer. The driver transformer is a Lundahl 1660-10. The driver stage operates at about 13 mA. 3 mA is provided by the feedback resistor, the remaining 10mA flows in the transformer. The screen grid of the driver stage is fed from regulated 150 volts, and the anode is at 200v (also regulated).

The filament of the output stage is driven from an isolated regulated DC supply. The connection to ground is through a 10 ohm resistor and that is used to set the bias. At 100mA quiescent, this is 1 volt. The stage is negative grid biased (about -60 volts). The screen grid is powered from a regulated 750 volts and the anode supply is powered from a regulated 800 volt supply. The output transformers are Lundahl 1620PP. Since the Matrix amp does not have uncancelled DC in either transformer, PP transformers can be used. As I mentioned, this circuit produces 66 watts (at about 5% distortion). With conventional SE transformers, this would be H E A V Y.

The parafeed choke is a Lundahl 1673-10H.

The 813 filaments draw pretty heavy current when first powered up. I used high quality switching supplies (12V, 6A) to power them. The initial load caused the power supply to go into current limit and fail to start. This was solved using low resistance inrush current limiters (thermistors). This also dropped the voltage perfectly to 10.0 volts across the filaments. The 6MB8 (or 6GX7) is also powered from these supplies, which keeps the input stage quiet. (Sometimes, an unbypassed cathode will pick up hum from AC powered heater. Powering from DC eliminates the problem).

The Lundahl output transformers with their undedicated 8 output windings are perfect for Matrix operation. For Matrix operation, effectively two windings are in series for each transformer (and two paralleled to them) for each channel. Then the effective "winding" for one transformer is connected in series aiding with the other transformer to derive left, and connected series opposing to derive the right channel. This also allows the two channels to be maintained independent and/or the "0 ohm" side grounded for both channels. This is not possible with "tapped" transformers (see the previous Matrix amplifiers). The symmetry also allows us to remove the resistor and capacitor that was in place on the secondary side of the previous Matrix amps.

The power supplies for this amp are ALL regulated. The master supply is the -100 volt regulator. That is, -100 volts is the REFERENCE for all the other output voltages, and these voltages track the -100 volt source. This was done so that in case the bias ever fails for whatever reason, the high voltages go away as well, saving the tubes. The -100 volt source itself is referenced from a sub-minature voltage reference (5783). A more common 5651 would do as well. Incidentally, sometimes VR tube firing voltage increases with no or low ambient light. I used a white LED optically coupled to the VR tube to prevent this.

The high voltage regulator is unconventional in that I used a light bulb as a non linear resistor instead of a heavier (and more expensive) choke. The regulators are separate for the screen grid and anode. The regulators use 6080 as pass with cascode 12AX7 for the control. The 12AX7 heater is referenced between +150 and +210V to prevent H-K breakdown, and the 6080 heaters are referenced to +750V to prevent H-K breakdown. Note the series connection of R3/R9 and R23/R24. Remember that resistors have a voltage as well as a power limitation.

There are separate +150, +200 and +210 volt regulators. The regulators use is shown in the table below:

Power Supplies
Voltage module Used for
-100V negative supply chassis bias, master reference for all other high voltages
bias1 negative supply chassis adjustable -40 to -80v for biasing one 813
bias2 negative supply chassis adjustable -40 to -80v for biasing the other 813
UM81 heater negative supply chassis 16.5 volt source to power the UM81 heaters
12V 6A #1 negative supply chassis filament for 813 and 6MB8(6GX7)
12V 6A #2 negative supply chassis filament for the other 813 and 6MB8(6GX7)
+800V positive supply chassis anode supply for 813. adjustable.
+750V positive supply chassis screen grid supply for the 813s
+210V positive supply chassis B+ for UM81, reference for 12AX7 in HV regs
+200V positive supply chassis B+ for 6MB8(6GX7) anode
+150V positive supply chassis screen grid supply for 6MB8(6GX7), reference for 12AX7 in HV regs

The time constants in the regulators are set so that power is applied in the following order:

  1. Heaters, filaments and -100V regulator available within 1 second, bias available.
  2. +210V and +200V in another few seconds
  3. +150V in another few seconds (note 150V can not be present without 210v - R20 in LV reg schematic)
  4. +800V after about 10 seconds
  5. +750V slightly after 800V (note that the 750V can not be present without the +800v - R14 in HV reg schematic)

On turn off, the bias is available for several seconds until the tube filaments have cooled off.


Sensitivity 0.4VRMS produces 10W into 8 ohms
Zout about 2 ohms
Freq Response -0.4dB at 10Hz, -0.1dB at 20Hz. Flat at 20kHz, +0.3dB at 30kHz, -3dB at 40kHz
Residual Noise about 600uV.
Power 66 watts per channel, both channels driven produces 5% distortion

Distortion (with 6MB8 driver) shown here:

Distortion with 6GX7 driver:

Note that the 6GX7 is indeed "cleaner". It also sounds slightly better, with no evidence of constriction at low levels. The 6GX7 change does not require any value change, but the socket connections are slightly different and are detailed in the table below (schematic shows 6MB8:

6GX7 and 6MB8 connections
Pin 6MB8 function 6GX7 function
1 triode grid (not used) cathode/g3 (note same as pin 3)
2 triode anode (not used) pentode g1
3 triode cathode (not used) cathode/g3 (note same as pin 1)
4 heater 1 heater 1
5 heater 2 heater 2
6 pentode anode pentode anode
7 pentode g2 pentode g2
8 pentode cathode/g3 triode anode (not used)
9 pentode g1 triode g1 (not used)

Thus, in changing to 6GX7, cathode connection moves from pin 8 to pins 1&3, g1 connection moves from pin 9 to pin 2.

Here's also a plot of distortion vs frequency and level for this amp with 6GX7 drivers:

At 20Hz and 60 watts, you're getting to transformer saturation.