EXTREME AUDIOPHILE CUSTOM AMPLIFIER
MADE OF PARALLEL MULTIBRAND OP-AMPs
<< Due to the intrinsic nature of the project
this amplifier is more "a custom" at amateur level
than professional and widely marketable.
I know quality is subjective,
but this amplifier is so exclusive,
it has such a solid build quality and it sounds so vivid, crystalline,
communicating an engaging soundstage,
that NOW I enjoy music like never before. >>
Copyright (c) by Gianluca G, Italy 2018-2021
IMPORTANT DISCLAIMER: today, 2018, manufacturer Linear Technology (LTC) is full part of Analog Device (ADI), but at the time of this design they were two distinct semiconductor companies, thus, I'm going here to mention them separately.
This is an independent hobbyst project and non-commercial site (https://ggianluca.wixsite.com/opamplifier): either LTC or ADI do not have any association with this work.
The information here presented is believed to be technically correct and everything presented on this site is done so in good faith. Anyhow you (the reader) are responsible for anything that you might do as a result of reading here.
I named the amplifier "The 50th Anniversary" because I built it as a gift to myself for my 50th birthday.
This project comes from the heart and its conception was simple: as I am a big enthusiast in my hobby audio applications of the tonal flavour of AD826 (ADI) and LT13xx (LTC) opamp family, why not build an amplifier powerful enough to drive loudspeakers by connecting many of these chips in parallel?
How would the sonic signature be mixing them together?
It is worth mentioning that none of these video opamps was designed for audio application by their manufacturers.
Sixty chips, $5 (in 2018) heterogeneous ADI and LTC dual video operational amplifiers in DIP8 package, were paralleled and work in unison in the audio band by sharing the same feedback in a master-slave configuration.
There are sixty opamps per audio channel, so 120 opamp in total, each capable of supplying 50mA individually, which can provide up to 3 Ampere current when paralleled. This is adequate to drive a pair of common 8-ohm loudspeakers at 10W power with minimal distortion and excellent linearity (0.003% THD @1Wrms; 0.006% IMD; ±0.1dB over 20-20kHz; >100dB SNR, what measured by me with non-professional PC sound card).
The amplifier is built on modular stacked high-quality fibreglass prototyping boards with multiple hole connections and power rail printed patterns and the circuit was designed with the aid of LTSpice, a freeware software circuit simulator created by LTC.
The idea is far from to be innovative as Douglas Self in 2010 already published a superb modular power amplifier in Elektor using a massive number of cheap NE5532 opamps in a grid configuration.
Before this, many semiconductor manufacturers proved in their datasheets (OPA111, LT6020, AD826 etc..) that it is safe to put in parallel ‘N’ stage of opamps to reduce the relative amplifier noise and increase the output driving capability.
But "The 50 Anniversary" differentiate itself by adopting a master-slave configuration. In this design a single opamp (the "master") feeds several other opamps (the "slaves") all placed within the master feedback loop and wired in parallel as unity gain current buffers replicating the master output voltage. With the master feedback being wrapped around the entire net of slaves, overall accuracy is maintained.
This approach is only effective at low frequencies such as audio, where the propagation delay of the slaves is negligible with respect to the wavelength of the input signal, so that the phase shift will not cause wasted current and distortion. You may find interesting readings about this topology in few application notes (like Intersil AN1111 or Apex AN26) and in various datasheets (like OPA511).
Because during the model simulation LTspice displayed excessive "ringing" at the output of square-wave pulses into capacitive loads, stability was improved by adding external compensation (a shunt capacitor across the master feedback resistor) small enough to don't jeopardise the high slew rates of these fast opamps.
"The 50 Anniversary" also combines mixed types of bipolar opamps (I cannot stand anymore Jfet opamp sound, sorry). If you listen carefully, you may perceive some tonal changes and psychoacoustic variations when connect identical or different opamps (either in series or in parallel).
My preferred opamp is the AD826 (ADI), which is smooth, open, clean and sounds true to life in mids and voices: this opamp reproduces the best female voice I've ever heard among any op amp tested.
To me also the LT13xx family of fast, high performance amplifiers employing Linear Technology Corporation’s advanced complementary bipolar processing (LT1352, LT1355, LT1358, LT1361 and LT1364) is audio superb!
The LT1358 (LTC) is warm, bright, tonal rich and sounds mid forward and plays with a slight exuberance but clean.
The LT1361 (LTC) is definitively one of the most crystal clean, neutral, and musical opamp I ever tested. It sounds detailed and sweet with full bass but it has not an emotional midrange impact as AD826 and LT1358 have: it is very good at input stages.
The LT1364 (LTC) is dynamic, fullness, punching bass, slight mid forward and sounds sweet but accurate.
Consequently, for the "master" amplifier it was chosen the AD826 (ADI) which alone pilots, at a gain of two, an equal mix of 59 (per channel) parallel AD826 (ADI) and LT1364 (LTC) in unity gain configuration. The pleasing parallel combination of the AD826 (ADI) with LT1364 (LTC) compensates for each other's sonic difference and exhibits the lowest listening fatiguing.
Both AD826 (50Mhz; 350 V/µs) and LT1364 (70Mhz; 1000V/µs) opamps exhibit outstanding technical specifications, quite difficult to find all in one sole amplifier: bipolar, single stage, handling all capacitive loads suited to drive cables, unity gain stable, hi-current output (50mA min), wide bandwidth (>50Mhz), extremely high slew rate, low-impedance driving capability (150 ohm), class AB output stage, easy to assemble in DIP package.
Also LT1208 (LTC) has stunning sonic and technical performances, comparable to AD826 (ADI), but it has half of its current capability, hence, it was here discarded.
The pre-amplifier section is made by one half of LT1361 (LTC) (in the 2021 enhanced version, which replaces the AD826 (ADI) of the previous version) as line pre amplifier and cable driver to the step attenuator volume control potentiometer, in series with one half of LT1361 (LTC) (in the 2021 enhanced version, which replaces the LT1358 (LTC) of the previous version) which provides most of the gain after the volume control: both are biased in class A by using Warren Young (tangentsoft.net/audio/) constant current source JFET cascade configuration. They have their own dedicated power supply using best-in-class complementary LT1963A/LT3015 (LTC) low noise low dropout regulators.
LT1361 not only sounds clean and accurate, but it also offers a low input offset voltage among bipolar opamps (1mV). This important parameter should be choosen as low as possible because this opamp is placed after the DC block capacitor and provides most of the amplifier gain: any DC imbalance in its inputs generates a large DC voltage at the output terminals of the amplifier which causes the flow of DC current through the amplifier and the loudspeaker, that wastes power, produces internal heat and prematurely cuts the output signal; it also increases speaker distortion, since the woofer's mechanical suspension would not be perfectly centered.
Two major enhancements were made in 2021 which have slightly lowered the measured level of Total Harmonic Distortion (THD) and Intermodulation Distortion (IMD) and also made the amplifier sound better to my ears, more accurate, sweet and smooth, but detailed.
First I added a second outer global negative feedback loop around the cascade driver stage and the power output stage (which already had its own inner feedback loop running among the master and unity gain slave devices).
The stability of such a complex nested multi-loop circuit has been intuitively determined with LTSPICE simulation checking the output response to a large signal square wave input and looking for the absence of possible oscillations, overshoot and ringing, unpredictable or unexpected response (also changing the load capacitance of the speaker model). Compensation is done either by the LTC / ADI op amp manufacturer inside the chip (all op amps are internally compensated for unity gain stability and can drive "unlimited" load capacitance while maintaining excellent phase margin) and also performed externally to the chips by putting a small capacitor (few pF) in each feedback loop, large enough to achieve the desired overshoot response without causing a loss of high frequency performance that would have impaired the sound stage aperture.
The main concern I had for this project was realizing a good cooling system because DIP packages are not intended to be mounted on heat sink at all.
Epoxy plastic is also not conductive: a DIP8 package has poor junction-to-ambient resistance, 130C°/W for LT1364 (little better in AD826), hence, without cooling system, it can dissipate only one watt at 20°C ambient temperature, reaching the maximum permissible internal temperature of 150°C.
All electrical circuits dissipate some power in the form of heat. Usually both LT1364 and AD826 run very warm even in standard application because their quiescent current is elevated (7.5mA per amplifier) being hi-speed bipolar devices: higher currents mean higher slew rates on internal nodes.
Heat wears out electronic components; as described by the Arrhenius effect equation, every 10°C increase in working temperature reduces component life by half.
The power dissipation on a single chip in this design could potentially reach nearly one watt (~500mW per opamp) in worst conditions, which I select in my calculations to occur driving a reactive 6 ohm load with 60° phase angle, like could be a tough home loudspeaker.
Calculations for heat dissipation in reactive loads are complex due to the phase difference between voltage and current, but Apex AN08 "Optimizing Output Power" application note helps with some ‘turnkey formulas' which I have expanded here. I considered a pure sine wave signal for simplicity, which is rather cautious as the long-term average level of audio tracks is much lower and depending on the type of music.
Power dissipated = PD = PI – PO where:
PI = power drawn from the power supply =
= DC power supply * AVERAGE output current + PQ =
= Vcc * 0.9 * Iout_RMS + PQ= 0.637 * Vcc * Vout_peak / |ZL|+ PQ
PQ = quiescent power supply =
= I_Quiescent * 2 * Vcc
PO = real power delivered to the load =
= Iout_RMS * Vout_RMS * cos(Φ) = 0.5 * Vout_peak^2 / |ZL| * cos(Φ)
|ZL| = 365 Ω, that is 6 Ω load shared among 60 opamp, plus Ro
|Ro| = 5 Ω, output resistor mandatory to prevent the parallel opamps from fighting each other
cos(Φ) = cos (60°) = 0.5
Vcc = ±15V
Vout_peak = max output swing before clipping = 13.7V
I_Quiescent = max supply current = 7.5mA
PI = 0.637 * 15 * 13.7/ 365 + PQ = 359mW + PQ
PQ = 7.4 * 2 * 15 = 222mW
PO = 0.5 * 13.7^2/ 365 * 0.5 = 129mW
PD = PI – PO = 452mW per opamp
Power dissipation per chip = 2 * PD = 904mW ~ 0.9W
Unbalanced output currents were not taken into consideration here.
Heat dissipation of 0.9W would rise the junction temperature to 117°C (=130*0.9) above ambient: because air temperature inside a cabinet might reach 47°C in summer, the chips will likely get damaged as their dies go above 150°C.
In fact, the indoor temperatures of houses in temperate zone could reach 35°C in July-August and the temperature rise inside a vented cabinet of a power amplifier rated like this, is approximately 12°C above external.
Hence, some effective cooling methods had to be adopted in this project.
First the cabinet has a fully vented top cover in substitution of the standard one and is equipped with high feet: this improves natural air convection moving hot air out from the top. The temperature rise, ΔT, inside the cabinet was reduced from 12° to about 10°C above external, internally reaching 45°C in hot summer.
Second, four heat sinks were screwed onto the perfboards, each covering fifteen DIP8 chips (or 30 opamps): the purpose of the sink is to remove heat from the top surface of the chips and distribute it to the ambient. Heat sinks were selected rather large offering a natural convection thermal resistance of 1.7 °C/W. But mounting them horizontally instead of vertically, like here, the natural airflow will be less effective, and the thermal resistance must be increased by about 15%, hence, to about 2°C/W.
Because fifteen devices dissipate 13.5W in total, we can expect the temperature of the heat sink (and consequently of top surfaces of the chips) to rise 27°C (=2*0.9*15) above ambient.
Given the temperature difference (per watt dissipated) between the die and the surface of a P-DIP8 device (called "Theta JC", 50°C/W for LT1364, which is in general about 2/5 of that to ambient), these chips would reach a maximum internal temperature of 117°C (=50*0.9+2*0.9*15+45) in hot summer and in worst power condition, at 0.9W.
Cooling multiple chips with a single heat sink is attractive but can create serious thermal issues: due to the differences in assembly of fifteen chips mounted in sockets, heat sink would not cover completely all chips because their top surfaces will be hardly coplanar. To prevent this, a thick (1.5 mm) and flexible thermal pad was inserted to fill any gap and ensure full contact between the heat sink and all fifteen devices.
Gap pads add extra thermal resistance between the chip and heat sink, which was minimized by using high performance silicone filled pad produced by Thermal Grizzly, ensuring a remarkable thermal conductivity of 8W/mk. This means that on top of a DIP8 surface area (60mm^2) the thermal resistance is only 3.2°C/W (=1000*1.5/ (60*8)).
Including the pad thermal resistance in above equation, junction would reach a temperature of 120°C (=50*0.9+2*0.9*15+3.2*0.9+45) when dissipating nearly one watt, which is safe but quite high, giving only 25 percent safety margin.
Because thermal performance of heat sinks can be doubled if put in airflow, internal fans were installed.
In general, adding fans to an audio amplifier is a bad idea, as their acoustical noise jeopardize sound performance at low listening levels.
This amplifier is equipped with two ultra-low-noise fans by Noctua, which run inaudible, at 9.5dB(A) combined, when underpowered.
The installed 80mm fans, which run at 8.5V at 1170 RPM generating 1.54m/s air velocity, halve the heat sink thermal resistance to 1°C/W. This gives extra temperature margin of 13.5°C: at 0.9W dissipation the die temperature will reach 106°C (=50*0.9+1*0.9*15+3.2*0.9+45) instead of 120°C, ensuring a conservative 40% safety margin in worst power and hot ambient conditions.
It is worth then recalling that a major heat flow path from the chip to the ambient would be through the board.
In a chip there are in fact two parallel paths of heat dissipation: part of the energy is carried away from the package surface and the remaining is conducted into the copper traces of the board through the device leads. Using sockets makes stuff worse as it adds another stage in the thermal path.
The choice to put extra layers of solder and to join large copper wires (17 AWG) along the traces, created extra paths in the board to remove heat from the chips as well as increased the current-carrying capability.
The surfaces of the four boards are mounted in the direction of the airflow of the fans and this makes the amplifier running cooler and more reliably.
During normal listening sessions the heat sinks work at 45°C (with 24°C indoor temperature), therefore each chip is dissipating 0.73W (=45-24-10/1/15), or 367mW per opamp, and the estimated junction temperature is 84°C (=45+3.2 *0.73+ 50*0.73).
For me the best-in-class amplifier is the one which presents the optimum compromise among the following criteria:
1) the "lower listening fatigue" after an extended period of time covering different music genres. The listening experience should be effortless for your brain: no need to guess or reconstruct missing information; nothing to add or to smooth out.
2) a "clean flat response" at any loudness level: you should not get an unpleasant audio experience while you turn up the volume. Colored amplification usually adds unnatural or annoying tones to the timbre, which often cause the desire to turn off the music and, nevertheless, never increase the volume.
3) a “vivid and emotional soundstage”, never flat or boring: you should not lose interest by continuously listening to your system. An amplifier with "0.001% distortion" must have its signature sound, whether you appreciate it or not, without necessarily being colourful because, soon or later, this will result fatiguing.
4) a “distinct difference" compared to ordinary products due to exclusive aesthetics and a peculiar design. As happens with luxury items, owning it must provide you with a unique sensory experience. The fact of having built, or even, designed it, offers you a deeper satisfaction.
"The 50th Anniversary" wins against all the machines I have tested, owned or built (even tube amplifiers!) because it offers the best trade-off among all above criteria.
Due to the intrinsic nature of the project, this amplifier is of course more "a custom" at amateur level than being professional and widely marketable.
In fact, it has:
- high parts count
- high parts cost
- low power per dollar ratio
- wasting space
- critical heat dissipation
But in the end, how does this amplifier sound?
I know quality is subjective, but to me this amplifier is so exclusive, it has such a solid build quality and it sounds so vivid, crystal-clear, communicating a full and engaging soundstage, that now I enjoy music like never before.
Last but not least, here are listed the remarkable features of this amplifier: independent left and right channels in a dual mono configuration; oversized transformer (400VA) for excellent regulation and current handling capability; dedicated windings for both power supply rails in each channel reducing crosstalk; Wima and Nichicon 105ºC capacitors (94.000µF in total); independent regulated low noise power supply for the preamplifier; 1% metal resistors; precise 24-step parallel attenuator (only two contact points in each position); star ground topology; source selector with distinct paths for hot and ground of the inputs; generous heat sinks; ultra silent Noctua fans with separate regulated power supply; solid chassis, feet, knobs and switch in full aluminium; gorgeous VU-meters tweaked with blue LED backlighting that give the amplifier a dynamic and unique aesthetic by mixing modern and vintage.
GianlucaG copyright, Italy 2018