|HotFET Pre - prototype|
The story has begun long ago with the perplexity caused by the fact that amplifiers based on vacuum tubes consume lots of power, get hot and do sound great; while the vast majority of silicon-based audio devices can stay cool and produce somewhat mediocre results. It did not feel right. Hence my attempt to bring some extra controversy into the world of high definition sound reproduction.
The basic idea of this design
The basic idea of this design was: let's put transistors into the "hot operation mode".
Let's agree on the terms: in this writing I'll utilize the word "transistor" to designate field effect transistors (FET) only. Bipolar junction transistors (BJT) are NOT considered here at all.
Below there are few fundamental thoughts that led to the design we are talking about here.
- Selecting the idle current in the most linear region on the device transfer curves is the common practice in valves. Nobody would regret that extra milliwatt dissipated at the tube's plate if it gives us less distortions at the end. In the silicon-based designs on contrary designers are trying to make it "green" and then fix all distortions by applying negative feedback loops afterwards.
- The higher is the bias current that flows through the device - the smaller fluctuations of it will be needed in order to attain the desired output signal swing. This way we utilize much smaller portion of the transfer curves of the device. Those small zones can be much better approximated by a straight line.
- The higher is the current that flows through a FET - the bigger its transconductance is, hence the amplification factor of a single device. This would allow us using less stages for obtaining desired amplification factor, or produce lower output impedance in the follower topologies.
All three above have one thing in common: they all strive to improve on an amplification device's linearity. That means lower distortions without any feedback applied. Negative feedback loops (NFL) are usually considered an asset in electronic designs. Only relatively recently advanced designers begun to realize that it can also be a liability. There are scientific proves of this statement (for ex. amps with deep NFL produce much higher numbers of harmonics in its distortion signatures). There are also not-so-scientifically-proven arguments to why not to use NFL in audio. In any case there's no such phenomenon as NFL at the sound frequencies in musical instruments, thus let's try to avoid using it altogether in the music reproduction chain as well.
There was much more to it, but currently I am struggling to recall all the bumps on the road that brought me to the current state of my thoughts. Few more assumptions taken into account in HotFET designs:
- Let's use n-channel FETs only. Their p-channel counterparts are in no exception two to three times worse in certain parameters, be it max power or capacitances, but at least some of those. It's being explained by lesser mobility of holes than electrons in semiconductors. Anyway there's no such thing as a positron-based thermionic valve yet. The same time there are great tube amps out there. Let's try to get away with only n-channel FETs too.
- They say "J-FETs sound better than MOSFETs". I personally would not sign under such a statement. However in this little design J-FET simply was the best fit based on its parameters and (I must admit 🙂 its tube-likelihood connection-wise. Depletion MOSFETs could also have been tried, but I do not seem to be able to obtain suitable devices of this kind in a decent quantity/price in any foreseen future.
- In tube amps there's usually a very generous headroom in power supply voltage (except for some output stages of the power amps).
The first real device
the first real device that has been built based on the philosophy sketched above was the preamplifier. To be precise it should rather be called an impedance matching device, or at least "the buffer". But it does have a volume knob on the front panel and those "in" and "out" RCA jacks on its back. It does not get really hot, but quite warm still. Hence - "HotFET Pre".
Choosing an amplification factor of this preamp was quite easy - it's slightly below 1. Most of modern CD players and other digital signal sources put out one or two Volts RMS. While power amplifiers are designed to produce full output being fed by 1 Volt RMS or even 0.775 Volt RMS (0dBm) at its inputs.
The working horse
The working horse in the HotFET Pre is the high frequency J-FET amplifier J310 (
So I really wanted to use it close to its Idss for best linearity. If we build a buffer for audio then we need to "repeat" say something around 4 Volts of amplitude. Thus the rough estimate of power dissipation per device: 30mA * 5V = 150mW. That's already on the borderline for devices in TO-92 case. And still we had almost no amplitude headroom. But look how sweet this mode could be: considering a "standard" power amplifier input impedance of 10KOhm we would need to pump some 0.4mA into such a load. That would let us with 1/75 of the bias current variations - the linearity dream that was.
Lately I was experimenting with the headphone amp based on this "hot philosophy". There it became evident that not only the bias current had to be high enough, but also the power supply voltage has to go sky high in order to obtain low distortion figures. Actually I was unable to trace any distortions (less than 0.01%) only once the power supply voltage went above twice the signal amplitude plus all the voltage drops in the circuit. (Just a reminder: there were no negative feedback loops.)
After all, shall we put all the desired load on our working pony J310 - it's getting sweet, very sweet, but way too hot. 10V and 30mA idle would yield 300mW. According to the data-sheet the device is rated at 625mW max. However taking into account the constant nature of the power we are going to put up and the thermal resistance from that very same data-sheet (357 °C/W) - we easily come to something above 130°C at the normal room temperature. What can be very comfortable for thermionic valves (130°C at its glass envelope) - can kill transistors.
The solution to that "too hot" issue was the good old cascode. The majority of the power dissipated will fall on a hefty MOSFET in TO-220 case, that could be mounted on a heat-sink. Minuscule J-FET will have to drop only about 1 Volt or so. As a bonus we reduce non-linearities caused by
As a load of our cathode-follower (oops, did I want to say source-follower here) we'll use an identical circuit.
|HotFET Pre - skeleton schematic|
The symmetry here is very beneficial in several aspects. The upper and lower arms will work in exactly same biasing conditions but in counter-phase while amplifying the signal. Thus reducing (mirroring) remaining non-linearities of each other. Provided J-FETs were closely matched and current setting resistors used (will be explained later) are of a high precision - the output offset is automatically set very close to zero and thermally compensated.
The necessity to select and match is good and bad at the same time. It certainly adds cost and complexity to the project. But on the other hand I do hope that this will at least delay those quirky solder-slingers from flooding online markets with re-branded semi-broken copies of my design.
I am considering to bring out to life a reasonably priced DIY kit with quality PCB and all critical parts (including matched J-FET quads) for building the real HotFET Pre (c) by those who are eager to listen to the highest quality music reproduction while still spending reasonable money and efforts. By the way please drop me a note should you be interested in purchasing such a kit - this may make it happen much sooner 😉
The next article goes in depth about this very simple schematic and provides sufficient details for those DIY folks willing to give it a try.