The idea behind this unit is to remove the sub-energy from the main amplifiers and loudspeakers. For a given voltage level, the lower the frequency the greater the magnetic flux in the signal transformers. So rather than implementing classic bi-amplification, I felt that it would be more effective sonically to remove the sub frequencies entirely from the main amplifiers and let the sub-woofer plate amplifier handle them. Similarly, for a given sound pressure level, the lower the frequency the greater the loudspeaker cone displacement. Consequently the sub frequencies cause doppler distortion, therefore there should be a real benefit in removing the sub energy from the main loudspeakers. I was astonished when I first tried this unit. There were clearly audible improvements in resolution and clarity but the biggest gains came with improvements to the sound stage, especially the tricky aspect of sound stage depth. Following the happy experience with the 50_300B constant-current topology, I recently modified this design to the same topology, each 2A3 is fed by a cascode mosfet constant current regulator to provide extreme isolation of the signal from the power supply. (For more infornation, see the notes at the foot of the 6AS7G PP project.)
The topology uses input transformers (Lundahl 1540) connected 1:1. Each secondary feeds a 2A3 grid. The 2A3s plates are choke fed with a potential divider across each choke to B+, the tap on each potential divider driving a 2nd order CL high pass filter, the final passband gain being around 6dB. The corner frequency of the high pass filters is around 50Hz. The series element of the potential divider appears in series with the plate resistance. Thus increase of Rp becomes a fraction of a larger resistance thereby reducing the effect of tube aging on the filter corner frequency. The shunt element further helps to swamp out the effects of tube aging since it appears in parallel with the plate and series resistance, thereby restoring the source impedance from the 2A3 output circuit to around 1kohm. Another benefit of the potential divider arrangement is that the loading of the 2A3 in the high-pass filter cut-off region is limited to around -4dB.
The high pass filter chokes are home made, wound on ferrite cores. I installed them into a copper box covered in self-adhesive mu-metal.
As with air cored coils, there is a critical trade between inductance and DCR. Initially, I wound up with 0.8H. Later, addeding some external iron magnetic circuit elements increased the inductance to 2.1H. There is a pocket of around 3/16" between each end of the cores and the walls of the box; I put a 1/16" build of EI lams (robbed from a wall wart transformer) in each pocket and wedged them hard against the ends of the ferrites using thin rubber. To minimize the crossover region loss, I arranged that the filter characteristic is close to Butterworth (flattest frequency response). I thought about going with a Bessel response (flattest phase response) but decided against it, having experienced some crossover region loss while experimenting with the concept. The filters terminate in a 5k switched series attenuator for an aggregate output impedance around from 1k minimum to 1.5k maximum.
The sub channels are buffered by separate emitter followers that have no filter, relying on the low pass filter on the sub woofer plate amplifier1. A nice benefit of using the 1540 transformers is that the units can accept both balanced and unbalanced inputs. (I use the XLR outputs from my Meridian 508-20 CD player also my DV563a player has XLR outputs.)
I chose 2A3s because I wanted to use a filamentary triode and needed a very low anode resistance. 300Bs would also be suitable. The 2A3s are great sounding but microphonic, and this did turn out to cause a problem. The Gemme audio Concerti 108 back-horn speakers I am using have a pronounced honk. After much frustration, I realised that the honk was interacting acoustically with the microphony of the 2A3s. Three silicone rings around each tube helped marginally, however what was needed was to improved the mechanical mounting of the tubes. The tube sockets are mounted on a turret board and I added some damping to the board by sandwiching a second layer made of what the Brits call hardboard using double-sided sticky foam mounting tape. At the same time, I took the opportunity to replace the ceramic tube sockets with teflon ones. Another line of attack on this issue was to reduce the loop gain of the system: Originally to permit control of all 4 outputs using a stereo control, I placed the attenuator across the secondary of the input transformer; Because this is before the gain stage (i.e. the 2A3), gain of the loop created by the 2A3 microphony feeding the line stage and power amp gain, feeding the speaker that is acoustically feeding back to the 2A3, was maximum at all listening levels. I rarely use anything like maximum level so it made sense to relocate the attenuator for the line outputs after the 2A3 so as to reduce the loop gain of the system at normal listening levels. These modifications produced the desired result, I am no longer (at last) perturbed by the horn honk issue!
To relocate the attenuator for the line outputs meant that I had to provide separate attenuators for the sub outputs and the line outputs. A Goldpoint 4 section Mini-V surface mount resistor series switched attenuator having two 20k sections and two 5k sections fitted the bill nicely. The two 20k sections are placed across the 1540 input transformer secondaries and feed the sub-woofer buffers. The two 5k sections are used to control the line outputs. This arrangement provides a nice rate of control and is extremely satisfactory. A nice benefit is that I can install (just) a Bent Audio remote drive to the level control at a later date.
Click to see the line stage schematic:
In line with what has become my standard practice for the heating of filamentary tubes, I use filament feed chokes and a current regulated supply. In this case, being a line stage and especially given that the volume control is on the input, filament supply noise suppression is critical. In a departure from my normal practice, to further reject line and rectification noise, I have incorporated voltage pre-regulators to the current regulated filament supplies
Click to see power supply schematic:
1Sub Woofer Channel Summation: Originally, I had the two sub channels connected to the two line inputs on the sub-woofer. While fooling around with the line amps off, I noticed a kind of 'cracking' distortion coming from the sub, akin to slew limiting on transients. I tried mixing the two sub outputs from the line stage using equal value resistors and feeding the result to one input on the sub-woofer, no better. Thinking (well perhaps not your actual thinking, more like muddling) about this, I came up with the notion that the two channels may contain fairly similar sub-bass signal content but at very different levels and also out-of-phase. So, I wondered if mixing the two signals in parallel was causing some form of destructive interference. What I wanted to do was to actually SUM the signals, not mix (muddle) them in parallel. What was needed was to feed each sub signal to separate transformers so as to isolate them from one another and then sum the signals by connecting the transformer secondaries in series. Since the sub-woofer has a plate amplifier and the sub paths in the line stage have less than unity gain, I tried using of all things, a (then redundant) pair of Lundahl LL9206 cartridge transformers, connected 2.5:1. The line stage sub outputs go to each "primary" while the secondaries are connected in series. Eureka! (Well I guess you may not.) No more cracking.