Maggie Mug: Active tube crossover for Magnepan 1.6

Author: Dmitry Nizhegorodov ( My other projects and articles


1.   In Brief

A two-way all-tube crossover for biamping of Magnepan 1.6 speakers. The crossover has low output impedance and is capable of driving difficult interconnects or amplifiers. The crossover effectively serves as a preamp, containing a gain stage, volume controls and input selector. The design is minimalistic, using total of four 6sn7 tubes. No feedback is used in the topology of the filters. The crossover was optimized using SPICE and then assembled on a chassis of Eico HF-60 amplifier.

2.   Design

Here is the design of the crossover, optimized in SPICE for best performance:


Only once stereo channel is shown. The input selector and the volume pot are not shown on the schematic, Also, not shown are the volume controls of the power gain stages that are part of the setup and are used as "balancing" controls.

Each filter is built from passive RC nodes. The filters are isolated from possibly reactant and impedance-varying loads by cathode followers. Additional followers also isolate the low pass filters, which are a difficult load, from the driver stage.

The topology offers good headroom and low distortion:

Top With good tubes, distortion can be under 0.1% with output swinging 5 VRMS into easy loads (tube amplifiers). Note that this is attenuated signal (~ -3dB, at crossover point). Full output amplitude for 3V of input amplitude is ~ 30V.

The crossover exhibits the familiar integration curve and offers flat amplitude response if the tweeter is inverted.

Top The combined amplitude response is the flat curve above. It is very close to staight line, with insignificant deviations (+- .1dB)

Top Phase shifts 90 degrees, as the sum of the 180 degrees shift of the low pass and 90 degrees shift of the inverted high pass signal. The extra math on the plot has to do with the result of the driving stage's 180 degree flip.

In case the tweeter adds its own phase and amplitude attenuation in the crossover region, the curve may bend slightly.

Top The combined amplitude response is the curve above. square is for tweeter's behaviour as 800 Hz high pass filter, then ~ 500 Hz filter, and so on. The tweeter attenuation was modeled by R9 set to 220 Ohm and C2 changing from 1.5n to 100n by 2 points per decade.

3.   Construction

After a round of SPICE optimizations, the crossover was built on a spare chassis for an Eico HF 60 amplifier.

Top The chassis allows to install 4 octal tubes and one 9-pin tube. One octal socket was needed for 5AR4 rectifier, hence one additional octal socket was installed. The chassis space that used to be the output Acrosound TO-330 is empty and can be used for future development.

Top Inside, point-to-point soldering was used with signal traveling clockwise from the input jacks (black panel RCAs) to the selector (top) to the volume control near it, to the two upper 6SN7s, each implementing a gain stage and a follower, through filters to the other two 6SN7s, one of which (middle) wired to contain two low-frequency output followers, and the other (leftmost) to contain two high-frequency output followers, then to the large orange output decoupling caps and bleeder resistors on the output RCA jacks.

Resistors are Allen Bradleys and garden variety carbon films. All signal path capacitors are polypropylene orange drops. Only .25mF were assembled for the low-frequency outputs, which is perfectly fine for any tube amp or any transistor amp with input impedance > 20k.

The power supply is tube rectified (5ar4) and heavily filtered. Heaters are AC-fed. See the next section for the details.

The power transformer is huge and is very excessive for the amp. It is believed to be genuine Eico HF60 power transformer.

One of the neatest properties of the transformer from EICO HF60 is 125V AC mains winding tap. Stock Eico HF60s are wired to the standard 115V tap, the 125v tap remaining unused. But what was normal in 1959 may not be OK in 2004. If you still read this you know the story yourself: not only most of us have much more than 115V in our wall power sockets, often 120 and more, but even if we have 115V, it seems to be too much for tube filaments, which glow too hot. What is going on? It could be that 110V was more common, and most transformers were optimized for that. When we plug vintage amps in AC lines with 120, 125 and sometimes even more voltage, we abuse the tubes. Thank to the 125V tap on HF-60's transformer, I was able to feed 6SN7s with ~6.1V and 5AR4 with 4.9 v - pretty much ideal.

4.   Troubleshooting, tweaking and listening impressions

The amplifier went up with all voltage and current readings as predicted by SPICE. Connecting the scope to the outputs revealed several problems, though. there was very severe parasitic oscillation and hum.

The oscillation was suppressed by means of 200 ohm resistors soldered directly onto the cathodes of the 4 output followers (originally all cathodes were wired to B+) Also, 1k resistors were installed at the grids of the gain tubes, as shown on the schematic.

After oscillation was tamed, a significant hum was revealed. Some improvements in ground wiring (move to star-ground-like wiring for the two groups of tubes did not help. The original design did not call for that, but a 15H choke and an additional capacitor were installed, forming a LCRCRC filter. The hum remained on the same level. Next was the time to look at the heaters supply. disconnecting it instantly removed the hum, hence the cause was revealed in the heater supply. The wiring was fine and a need to go DC was almost apparent. However, the old trick of lifting the heaters potential was tried, and that killed all of the 120HZ hum. Beneath, there were classical parasitic capacitance-induced short and sharp peaks, spaced at 60HZ. Most likely, the transformer from HF-60 was not perfectly wound or it is simply too big for the job. The spikes were suppressed by .1mF capacitor connecting the heaters to the ground.

After that, it was time to listen to the crossover. The sound was smooth and open and the background was quiet.

With 20dB of headroom, and low impedance, driving abilities are very good, and this results in punchy, enjoyable sound.

The scope shows that the crossover can swing out up to 20 V RMS before visible distortion kicks in, and this distortion is due to the input stage running out of its class A region into zero-bias. Clearly, 20 VRMS is very good overhead clearance, whereas the target output window is 1V RMS (e.g it is 26dB clearance). At this level, distortion measured with Russian 6H8C tubes was very good, and even 3V out gave around 0.1% THD. Actual distortion figures vary somewhat with tubes in used. The spectrum is very much SE-like, with 2nd harmonic dominating.

Top Signal spectrum for 200 Hz Sine wave signal, 1 V RMS output. Low-frequency channel.

Top Signal spectrum for 200 Hz Sine wave signal, 3 V RMS output. Low-frequency channel.

Top Signal spectrum for 2000 Hz Sine wave signal, 1 V RMS output. High-frequency channel.

Channel to channel comparison. I ran the scope in XY mode, comparing the left and right channel output of both filters. In this mode, the scope is supposed to show a straight line at ~45 degrees, the angle should not change through the frequency range and the line should not turn into an ellipse. The line remained straight and the angle did not change. This means channel to channel attenuation accuracy < .1 dB and phase balance < 1 degree.

Integration. In scope tests, the combined response was very much like what is predicted by SPICE, see the amplitude plot above. Note: one of the signals *must* be inverted for this test. For example, this can be tested with a dual trace scope in ADD mode and with one channel inverted. If you forget to invert, it will result in 6..8 dB notch in the crossover region.

5.   Discussion, background, design decisions

Active biampling of Magnepan 1.6 speakers, a very desirable arrangement among Magnepan enthusiasts, calls for a low-signal 2-way crossover. Since all-tube amplification rig is assumed, a tube crossover was contemplated. The crossover point for Magnepan 1.6 is 600Hz. After several iterations in preliminary thinking phase, a topology based on a gain stage feeding two filters surrounded by cathode followers was selected as the base candidate.

Why tubes were needed at all, and can the tube count be reduced? It is tempting to build single-tube filters, as unity-gain cathode followers, such as [4], and even more tempting to build the crossover as a completely passive device, see for example [1], hoping small insertion loss is OK in the world of CD/DVD players swinging up to 2V.

Why the output followers. Without the output followers, the crossover would be able to drive loads of only specific impedance and would be highly picky about interconnects. A gain stage with a line output transformer is an alternative to the output cathode followers, and it was considered as a possible future experiment direction. Yet the followers were chosen for simplicity, clean sound and excellent driving capabilities. Cathode followers must never be pushed towards their amplitude envelope limits, and then they sound very open.

Why the input tubes, and followers on the low pass. Unless there is a preamp with gain control upstream, the crossover needs volume potentiometers. Feeding the filters from a volume pot will distort integration, unless its impedance is very low, hence a buffering stage is needed, and in our case, this stage is also a gain stage. Without the input cathode followers, the low pass filters would require a source which impedance is low and constant, otherwise that would disturb the time constant. The high pass filter is a relatively easy load but also is disturbed by source impedance varying widely, and hence must be isolated from the volume control, but need not be attached to the input follower. In fact, taking the high frequency signal from the follower will increase distortion for the high frequencies due to shunting nature of the low-frequency filter.

Since the stock Magnepan 1.6 contains a 2nd order low-frequency filter and a first order high-frequency filter, and the same seems to be the configuration for active filtering used by many Magnepan enthusiasts, same was selected for initial design. At first, an all-passive low-signal filter similar to described in [1] was tried. Here is its SPICE results:

Top passive signal line crossover, optimized for amps with 100K input impedance and allowing to use a volume pot, with low 5k resistance. Higher values influence the filters much more.


What order? One of the requirements was to preserve a possibility to experiment with all-first-order or all-second-order crossovers. It is known that all-first-order crossovers offers the best integration - flat amplitude and phase. All-second-order crossovers offer only flat amplitude, not phase, and require signal inversion, otherwise severe notching occurs. Why not to use one of these in the initial design? The stock Magnepan 1.6 contains a second order filter for its low-frequency panel. Its high-frequency panel is inverted, and connected via a capacitor, forming a first-order filter, it looks like it adopts the all-second-order scheme, with inverted high-frequency signal, except one pole is missing. The answer may be in the nature of Magnepan's construction. Its high-frequency transducer is a planar (ok, "quasi-ribbon", to be exact) panel which does not need be crossed to limit its bandwidth; it appears to have a natural baffle roll-off by itself and that roll-off represents a first-order acoustical filter.

Top Measurments, see [3] for the complete picture, support this, clearly showing 12 dB/oct slope for the tweeter fed by a 6db slope HF filter.

My own measurments of Magnepan speakers (biamped, via passive low signal circuit, the mock/prototype for this tube variant, with pink noise as input) demonstrated fairly flat, notch-free spectrum. If this all is correct, then the crossover indeed needs a first-order filter and will indeed integrate well with a second order low-frequency filter, connected out of phase. Yet a possibility to experiment either with a first-order filter at the bottom or with a second order filter at the top seems very attractive, and the design allows this possibility.

6.   Variations and Possible Future Directions

As a variation, the low-frequency channel can be turned into a first-order filter. It is very interesting to try that on Magnepan 1.6 speakers. This can be easily done, even as a switch. the idea is to short R6 and to disconnect C5. There is even a simpler, and perhaps, better approach - R6 gets shorted and C5 stays in place. the result is slightly lower RC filter's F3, but that may be OK.

Another variation is to convert the high-frequency filter to 2nd order. Going under 1nF is impractical, hence either a low impedance RC node is installed in front of C7, and that is connected to the driver follower's cathode, or a Sallen-Key circuit is used. The latter allows to avoid feeding the high-frequency circuit from the cathode of the follower shunted at high-frequency by the low-frequency circuit's R1.

John Broskie suggested using a bipolar power supply in cathode-follower-based filters, which promises better PSRR and also helps to simplify biasing arrangements in the high-frequency path. Converting a full-wave centertapped tube rectifier to a bipolar supply is very unorthodox, but it can be tried as an interesting experiment.

The followers can be replaced with SRPP, White followers or Broskie followers.

Yet another improvement may be in optimizing the output followers for 50% lower swing, but more current. This will call for reduction of the plate voltage and more current.

7.   References





Author: Dmitry Nizhegorodov ( My other projects and articles