This description helps to lock down the designs of the low-level circuits, settle down their interactions, and decide the power supplies.
Refer to the table in web page https://sites.google.com/site/solderandcircuits/home/teachers--topics-in-common-with-your-teaching/big-speaker-idea-for-2016-career-day, which is a flow chart titled Custom Audio Amplifier Outline. The green-background "module" in the flow chart is the low-level circuits. Also on the low-level PCB are the row-Mic and row-Sine circuits. These circuits are being designed this week, March 16, 2015. The project's overall success, the flexibility of the system, & the level of excuses, if any, depend as much on the low-level circuits as on the high-power amplifiers.
At cell F/Woof, the left-right stereo mixer is an op-amp summing circuit to get any stereo signal down to monaural. There is some EMI filter but not as much as Input/Woof. The F/Woof mixer input is only 3.5mm jack. Two switches admit or block the left & right channels. The output is only phono jack (RCA) & may be patched to the Input/Woof input. Gain is one.
The J/Woof soft clipper is being moved into Amp/Woof, & will be inserted after some of the gain of 41 in that cell.
The Input/Sine reference oscillator at 65.4Hz, with low harmonics, is from the excellent op-amp book, Op Amps Design & Applications, 1971, McGraw-Hill. The choice of circuit is one I haven't tried before, it is the quadrature oscillator with nonlinear amplitude limiting. Since I already have CMOS 4066 analog switches, I am doing amplitude limiting in a fancy way, not so much to improve performance but to see how well the fancy way works, & whether logic can assist analog. There is a sine zero-cross (+ slope) sensor consisting of comparators F & G. At a quarter cycle later, 3.8ms as timed by a CMOS 555, comparator E determines if the sine amplitude is short of 5.5V peak or over 5.5V. The analog switch will either connect approx. 1Mohm across the second op-amp's integrator cap (to start reducing amplitude) or not do that, to build up amplitude. The application of the 1M will last a full cycle of 65.4Hz. A seven-state state machine manages the actions. (The operation of the state machine makes a good story for school students, an easy-to-follow sequential logic circuit.) Two LEDs indicate duty cycle of analog switch on & off. The Atmel CPLD will also have a 100kHz clock divider, down to 1Hz, for any control-panel lamp flashing that may be needed. In the loop of the oscillator, there are two pots to trim R1C1, to get a tiny bit higher than R2C2. The fine tune is a 4mm pot, SMD. It needs to be at a convenient, exposed point on the PCB. (It will not be a pot with a knob on the control panel.) The reset button for the state machine, which may never be needed, should be near the fine-trim pot.
The .3V peak LED at cell F/Woof is really two LEDs, a low-freq LED and a high-freq LED, approx. 3Hz to 200Hz and 200Hz to 3kHz. These are aids to setting amplitude of the "attenuate or amplify" at cell F/Woof, but accuracy isn't afforded by them, they are just rough aids.
The Input/Mic amplifier is for dynamic or self-powered condenser microphone with low-impedance, balanced, XLR cable to the amplifier's connector, a male-pin XLR receptacle. There is a lot of anti-EMI, starting on the solder pins of the receptacle. Frequency range is 5Hz to 10kHz. Overall gain is 800, apportioned as 33 in the BJT diff amp and 24 in the diff-to-single-ended op amp circuit. The latter can be cut back to gain of 3.1 (overall gain 102). The first-stage, gain-related resistors are low-noise, thin-film, SMD resistors. The first-stage diff amp is buffered by emitter followers on the way through capacitor coupling to the op amps. There are RCA and BNC output jacks. This XLR amplifier is intended to be a low-noise, high quality monitor amplifier that is useful to critically evaluate distortion and hum in the other circuits of the project. The layout of the Input/Mic amplifier must be with attention to oscillation susceptibility, since the +-23V output of the subwoofer amplifier is close by. Inadvertent feedback of only one part in 50,000 can cause the system to oscillate. This mic amp follows on two previous mic amps I have done, which were quite successful and had low susceptibility to radio stations. Full advantage is taken with the Input/Mic amplifier of the +-24V supplies, & those supplies are heavily filtered for use by the mic amp. (The Input/Mic amp could be designed for low-power 9V operation, but +-24V are available & it makes the circuit design easier.)
To complement the Input/Mic amplifier, a headphone amplifier with low output swing and low output impedance is provided. It is gain 0.9. It is Class A. The input is RCA, output is monaural phone 1/4" with a switch to optionally parallel the stereo channels of headphones. The output connector is wired with a common-mode choke to try to keep this high-current amp from feeding back to the microphone-amp input, which would cause oscillation. The supplies for this amp are +7.5V & -5.15V, which are surplus wall-pack supplies, AC to DC. Current in the Class A transistor is nominally .3A. Class A amplifiers that can drive low impedance loads, like 16 ohm headphone or 8 ohm speaker, always have heat. The Class A transistor must be heat sinked for 4W. There is a volume control in this amplifier to assist in keeping it from clipping, a common problem with low-impedance, low-voltage-supply amplification. The little wall-pack supplies must be located away from amplifiers because there may be high 60Hz magnetic fields. Grounding of the output 1/4" phone jack must be through the common-mode choke mentioned above, then directly to the +- electrolytic decoupling caps, and then to star ground of the project. The two transistors in the amplifier are PNP emitter follower followed by NPN emitter follower. They are arranged to have close-to-ground DC bias at the headphone output, for low click upon pushing the headphone's phone plug into the output jack.
The subwoofer-cone-resonance, bandstop filter at P/Woof in the flow chart is intended to keep the subwoofer's cone travel from becoming excessive at about 24.4Hz. What is not known is whether excessive group delay is introduced by the bandstop filter. The bandstop filter was designed on March 15, but it has to be built and there will be a lot of adjustment. This bandstop filter is important to getting flat response from sub-audible frequency (like 1Hz) to 80Hz. The potentiometer adjustments for center frequency and Q need to be easy to reach, not buried in the middle of the PCB.
The electronic crossovers at M/Woof and M/Mid are not designed as of March 18. The aim of making the crossover freq switch adjustable, like at 80 Hz or 160Hz, means there will be multiple filters, or analog switches will be used to adjust resistors.
Series regulators: +-6V for 4066 are on the green-background board, as are 5V logic regulator & regulator from 16V to 12V for relays. 5V regulator is from a surplus 7.5V 2.1A wall adapter that also supplies up to .3A to the headphone monitor amplifier, a Class-A amplifier. The 7.5V & -5.15V wall packs are packaged with the +-24V 8.3A, open-frame supplies because these four supplies need 120VAC input. Supply grounding must be carefully star-grounded, & the grounding arrangements must be given room, not haphazardly tacked together. There is even need to consider magnetic field around the highest-current grounding wires or straps, & not let the field cause voltage generation onto other ground wires or circuits.
Progress March 24: most low-level circuits are designed. I have taken stock of volume controls & switches in preparation for parts & supply orders. The layout of the complex front panel is mostly done, with attention to keeping the wiring from getting too messy, & to make the project free of self-oscillation. The front panel will be aluminum, of sufficient thickness to let 1/4" phone plug for headphone be inserted without bending, but there will be a lot of holes for the pots & switches.
While reading in Bob Cordell's web site, I see reference to the new, low-noise, dual-JFET, integrated circuit LSK489 by Linear Integrated Systems & available probably at Round Rock, TX, distributor Trendsetter around $6. These monolithic JFETs have Vgs matching around 20mV & much better temp match than the SMD discrete JFETs I was going to use for my 1/f noise measuring project.
March 30, 2015: While reading Bob Cordell's Designing Audio Power Amplifiers and reading in his web site, I see that better integrated circuits would be good for this project. Some of these are Analog Devices designs which were bought up by Texas Instruments.
1) LM3886 is a speaker amplifier with protection. It can put tens of watts onto speakers at THD around .03%, which isn't great but probably adequate for the midrange amplifier, above 80 or 140Hz. It has 11 pins & bolts onto a heat sink. Mute input. $6. Bought some & have one working in the midrange-strobe PCB, Sept 2015.
2) OPA551FAKTWT is a power op amp for $7. The package is DDPAK which is surface mount and can solder to a copper or brass heat sink. +-30V supplies, far beyond standard +-15V. It is stable to gain of one, 3MHz GBW, 15V/us slew rate. (If you don't need so low a gain, 552 is stable down to gain of 5, 12MHz GBW, 24V/us.) With a big heat sink, this power op amp has thermal resistance down to 25degC/watt. This op amp would be ideal for the fixed gain block of the subwoofer amplifier, which takes line level (like .3V peak) up to +-12.3V pk (bridged amplifier), which is gain of 41. This is working, with soft clipping, as of August.
3) The venerable 4-quadrant analog multiplier chip is still marketed. From Analog Devices/TI, it is MPY634 & it is surely laser trimmed on-chip for accuracy. But the price is $24, so I probably won't purchase this one. AD633 is $8, better. I need multiplication of speaker voltage & current to get power in the voice coil, including non-sines & arbitrary phase due to cone resonance. (Speaker efficiency is always very low, consider that the acoustic power leaving the cone is so small that all the power flowing into coil is lost as heat.) There is a "triangle averaging multiplier" in Op Amps Design & Applications that can handle up to 80Hz and 140Hz; elaborate but cheap. This is a sub-project that needs to wait for the bulk of the amplifiers to be proven, but build in a D'Arsonval meter on the front panel so metering of the speaker power can be added later. Other possibilities for multiplying are multiplying DAC, use of 3-bit ADCs to get digital V & I, then go into PROM or Atmel CPLD & look up product. If PROM, then probably 4-bit V & I. This is pretty easy to do (couple hours design) at 4-bit product or 5-bit. A really neat option is to use this technique for safe operating area protection for output transistors, though on April 2 it looks like SOA is inherently safe . The speed of this technique means one CPLD might handle multiple power/SOA calculations, with analog switches funneling different variables in. Just one ADC could work, CPLD could latch V while ADC looks at I. Arduino Uno is another possibility, which might work for both woofer & midrange. But Mega2560 may be needed for it's I/O qty.
Instead of using multiplier to find speaker power, I will use the left 4uH, wound with thinner wire, to simulate the voice-coil power, sensing with a thermistor. The thermistor will feed into the front-panel meter to give an indication of voice-coil temperature.