JLH gets regulated (temporarily)...

Over the weekend, I tried to get modify some of the high efficiency DC/DC converter boards to output 26V for the JLH, by replacing most of the passive components with different values suggested by TI's design software, but not a great deal of joy was had - while they will happily do 26V into no load, the under voltage lockout/ramping doesn't appear to be right as they only output around 3.3V when connected up to the amp... ah well.

So why not try plugging into the linear bench supply?  If memory serves (I built it a few years ago), this is LT1084 based, so capable of decent grunt with the die cast aluminium case acting as a heatsink.  The outputs have been configured for a voltage of around 33V.

First results are promising with an 8 ohm load - modulation is now down to almost nothing, and the noise floor is much flatter... but there's a lot of high order harmonics present that weren't seen before in the unregulated supply.  Let's try reintroducing that hand-wound 2mH 0.03 ohm choke between the supply and the JLH... hey presto, much cleaner!

(Note: signals are still being normalised to 0dBFS)

It's important to remember that most linear regulator ICs are not good at suppressing higher frequency noise... that said, it is surprising just how big an effect the choke is having.  The performance is now very impressive indeed for a simple circuit, and fully satisfies a basic objective criteria for a "clean watt".

Let's go further and try it with a 15 ohm load again, this time comparing to the measurement reference DAC being used as a source in these tests.  The DAC is based around an old WM8740 evaluation board whose heart has been replaced by an AK4396 in the name of better objective measurements... here's what it looks like inside...

All the mains circuitry in the picture is purely for charging - when in use, the design is disconnected from the mains and the DAC runs fully off battery power, with digital input coming via. optical for electrical isolation - this eliminates the possibility of annoying ground loops at the input end.

The measurement reference DAC is driven with a 24-bit 96kHz test signal, captured by a modified LynxTWO-B board with AK5394 A/D conversion... this is as high as the board will take in its current form via. S/PDIF, and appears to give good results... while it is certainly possible to achieve lower distortion than this (the simplest means being high order filtering around the test tones), the results are plenty good enough for a basic check point.

Putting to one side the second and third order harmonic, the JLH closely mirrors the distortion and noise of the input signal... in fact, if anything, you could say that the JLH appears to have "cleaned" the signal up a bit - this suggests that the output of the DAC probably would benefit from a little bit more filtering.

This is pretty good performance and suggests I should build up some linear regulators as soon as possible for the JLH so I can go back and listen to them again!  While a discrete regulator would be interesting to play with, I've ordered some LT1083 to do an initial first run with, as the design seems to work so well with even a traditional series regulator.

Class A contender no.1 - JLH 1969

When it comes to audio, in my opinion, quite often simple is good, sometimes best.  It's very easy to overcomplicate designs and introduce additional problems, only having to add additional parts to ameliorate basic flaws.

A perfect amplifier, as someone once said, is a piece of straight wire with gain.  While there are plenty of line level ICs such as opamps capable of vanishingly small levels of measurable distortion, achieving the same with much higher levels of current and voltage is much more taxing.

As has been said, running in Class A gives a head start in terms of low distortion, as it avoids the problem of crossover distortion in Class B amplifiers, where the amplifier switches fully from sourcing to sinking current, or vice versa.  To lessen the problem, Class B amplifiers are often offset biased to become a Class AB, so that the amplifier operates effectively in Class A for a fraction of its rated power.  This is very effective at reducing crossover distortion, but as Class AB amplifiers tend to be designed for much higher powers (and therefore, typically run on higher rail voltages), the bias amount tends to be fairly small to keep consumption down.

For the sake of a single clean audio watt, I believe that (with mains power at least!) Class A is probably the best way of achieving the goal... as an added benefit, Class A circuitry tends to be simpler which means from my perspective, less componentry to sully the sound.  An excellent reference on the subject of Class A amplifiers is the Class A amplifier site, run by Geoff Moss.  The site focuses on variations of an amplifier developed by John Linsley Hood, or JLH as he is often referred to.

JLH came up with quite a brilliant little Class A design which was the subject of a Wireless World article back in 1969, and is still a reference today.  It uses just four transistors, one input level PNP, a mid powered NPN for phase splitting and two beefy NPN power transistors for the output.  As was common for the time, this is a single rail non-complementary design which means you can really go to town on just one rail of power supply and not worry about how complementary your output pair really is.

One of the reasons why the design needs so few active parts and can run single rail is due to the simplistic biasing and that both the input and output are AC coupled (so coupling capacitors in the signal path).  Some audio enthusiasts run a mile at the thought of capacitors in the signal path, but I am not one of them.  Providing the cap is of high quality, I'll happily take one over a far less linear active device which will leave a much bigger sonic imprint (to my ears, anyway).  They also provide a degree of safety over that inevitable time when then the bias "wanders" and there is an unpleasant amount of DC going where it is not desired...

I won't go into the circuit operation in detail as far wiser heads than me have debated it in great detail over the years, but it can be read as a simple three stage amplifier.  While I've grown to have a fondness for FETs over the past few years, the sound of a JLH amplifier has always stuck with me, so thought it was an excellent bipolar-based design to start with.

As someone who's designed a few bits of audio kit in their time (and still do, when time permits!), it feels a little lazy to use someone elses' PCB, but you have to value your time when it comes to these things... there is often little point in reinventing the wheel, particularly when going after a fairly faithful recreation.

I decided to start with the first iteration of the JLH design, as this is the simplest (and you could argue, purest) form of a bipolar Class A that you could wish for.  I found a seller on eBay that does what appeared to be authentic looking 1969-design PCBs with the added bonus of supporting more modern component pinouts if you wanted to try different parts.

On the whole, I was keen to stick close to the original transistors, with the exception of the output transistors... by all accounts, the OnSemi MJE15003 are considerably superior to the originals in this design and so could be used without hesitation.  TO-3 packages are a pain to mount compared to the more modern TO-247/TO-3P as they usually need an angle bracket when using with a PCB... this is then thermally coupled to the primary heatsink.

For extra security against misalignment, I put PTFE sleeving around the TO-3 pins... this ensures even if the TO-3 packages somehow wiggle their way to making contact with the metal, no shorting should occur.  Fancy alumina ceramic shims (about 1.5mm thick or so) were used to thermally couple but electrically isolate the TO-3s from the heatsink, with liberal use of good quality thermal paste in the sandwich. This naturally gives a bit more lead inductance but reduces stray capacitance.

I chose to keep the original 2N3906 PNP (using a Magnatec part which should be very close as a second source to the original Motorola version) for the input transistor, though for the NPN splitter, I ended up going with a 2N1711 branded part with lovely-looking gold plated leads - the 2N1711 was endorsed by JLH as a superior replacement so feels authentic enough.  All transistors were tightly Hfe matched, across both pairs and channels.  I would have preferred higher Hfe parts but out of 16 power transistors, the "best" were around 50ish, with a lot of them much lower.

There didn't seem much point to throw exotica at this first attempt, so you won't see any teflon capacitors or tantalum resistors here - the input cap is a salvaged WIMA Polypropylene from an amplifier refresh, and the output capacitors are effectively "no name" Forever-branded units of basic merit.  The decoupling capacitors are good quality Rubycons, bought in for the job

 I socketed R5 in the picture so that I could tune the output to be half the rail voltage as JLH recommends, but the fixed 100K was so close to half rail already that I took the trimmer pot out and put the 100K back in.

For the sake of getting things up and running quickly, I wanted to skip the regulated supply and try one of the many simple unregulated supplies floating around here of many voltages - unfortunately never quite the *right* voltages, it seems!  Commandeering a set of 2x25V 160VA toroids from Antrim, back when Maplin used to sell more interesting componentry, I put the secondaries in parallel, rigged up a simple full wave rectifier and threw a big Elna Cerafine on the output.  Hm... a bit high... off-load DC voltage was something like 45V!

This design is meant to run off 27V for 8 ohm loads... more could certainly be tolerated by the parts in question, but that it going to get properly toasty at that, never mind being worried about the health of the some of the parts from a voltage point of view.  A bit of thought, and I remembered that there were a couple of 100VA toroid cores spare which I'd intended to wind chokes with... a spool of 30A wire and a patient Anna resulted in two simple chokes of approximately 2mH each, and very low DCR.  These were put in series with the rectifier, and kept well away from the mains toroid in use.

As the angle brackets were rather oversized for the job, it was thought that it would function as a basic heatsink for now - after all, it should be only 30W or so per channel.

Both amplifiers were gingerly powered up, and gladly showed signs of sane biasing the first time around, rail being between approximately 32 and 35V.  The bias starts off fairly low and stabilises at a considerably higher point, being similar for both channels... the bias appears to be very sensitive indeed to temperature - just grabbing the heatsink with your hand is enough to affect the bias to a significant degree, which does sound like an element of the design that will benefit from slightly more complexity!

Ok... enough of this faffing about, let's get them into the main system and see how they sound.  First night impressions were very positive indeed... while I wouldn't call the resulting sound "airy", it certainly came across as more beguiling and of fine definition.  I've experimented with many amplifiers over the years but have usually come back to my humble Arcam Alpha 8Ps... a quite traditional (and relatively complex) Class AB amplifier with a complementary MOSFET output stage - the JLH was certainly bringing something new, though quite hard to define.

Let's see what some measurements show.  For sake of brevity, I'm just going to show some normalised 24/96 65536-point FFTs of a 1kHz input... load is a 25W wirewound power resistor of either 8.2 or 15 ohms.  This lets you see the harmonic spectrum, and give a great deal more information that any single THD figure will.  While these traditional measurements only give, IMO, a small insight into the sound quality of a device, the order and shape of the harmonics can be quite revealing.  There have been alternative tests proposed for quite some years, but this a reasonable starting point.

Let's start with a 15 ohm load first.  This was done at considerably less than a watt as the MF+HF units will rarely get anywhere near a full watt... if they do for any length of time, I'll probably have my fingers in my ears, and perhaps the neighbours might want a word...

The first thing that is immediately obvious is that the JLH output is being modulated, probably from mains harmonics... at a relatively low level, but nevertheless impacting on the sound.  This is likely to be a combination of the very primitive unregulated supply and the primitive biasing arrangement.

The second thing that struck me is how much (relatively speaking) high order harmonic distortion the Arcam has.  While this amplifier will no doubt measure very well in terms of a THD figure, the harmonics of this single test tone are spread across the whole frequency range.  Second order harmonic distortion in my experience is fairly benign and is generally overwhelmed by the speaker contribution, so isn't worth worrying about unless of a very high quantity... third order is a bit more concerning, and I'll generally like to see it below -80dB on the reproduction front, and fifth and above odd harmonics preferably below the noise floor.

A high noise floor is evident on the Arcam, possibly down in part due to the high gain that this amplifier offers, considerably higher than that of the JLH - I suspect when the JLH is given a better power supply and better biasing, the noise floor will drop further to go with the lower modulation.

Let's have a look at an 8.2 ohm load now, at close to a watt - this is considerably harder than I'm putting on the amplifier at the moment due to a resistive attenuator network for gain matching, but is interesting for comparison purposes.

Note that with the more difficult (albeit still only mostly resistive) load, the modulation on the JLH output has dropped considerably.  The JLH second harmonic is quite high (which would lend a rather poor THD score, for what that's worth - not very much) and slightly higher on third harmonic, but aside from a few odd glitches, higher harmonics are pretty much absent.  Odd order harmonics on the Arcam are visible all the way up the 15th, and this isn't even with extra averaging.

So the Arcam isn't very good objectively at low power levels... consistent, yes, but not particularly low in even basic distortion tests.  The JLH is already sounding good, although these results do suggest that the modulation will be impacting on the sound - hard to know if this is being perceived positively or negatively at the moment.

In any case, there is more work to do on the JLH... whether I will attempt to modify the existing boards or start afresh, I'm not sure, but it's certainly worth some more listening hours!  :)

The perfect watt...

And so we move neatly to the subject of amplification.  The amplifier in a system is generally dictated by the speakers, as different speakers have different requirements.

Broadly speaking, the majority of modern commercial speakers are in the region of 86-90dB/W/m, which is what I'd class as "mid efficiency" units.  These can be driven to modest levels by almost any amplifier, but ideally 20 watts or more, depending on how wild the impedance curve is.

Heading into the realm of drive units designed for high SPLs (a typical example would be for PA use) are units in region of 95-100dB/W/m... I class these as "high efficiency" drive units, which typically have very powerful magnet systems and lighter diaphragms.  There is a penalty (other than the typically high price) in that the lighter cones often lack the critical damping of lower efficiency units, which can lead to considerable colouration in the sound... however, the gains can be worth it.

Once you are beyond 100dB/W/m, you are in "super high efficiency" territory.  While even a 100dB drive unit may only be technically a few percent efficient in terms of converting electrical power into sound, these units are vastly more efficient than a conventional unit due to the log scaling.  The use of horns in particular can allow (given enough space!) very high SPLs from only a handful of watts.  If you are prepared to spend serious money, then 110dB/w is feasible!  At this point, an amplifier is practically unnecessary - a liability, even.

Back in the real world, we have limited resources, and without the space for large horns (or wanting to deal with their own set of issues), "high" efficiency is a reasonable goal.

My own speakers are essentially divided into two.  A bass driver operating essentially in free air, which is very inefficient, and a midrange and HF unit of relatively high efficiency.  While the bass driver requires a powerful amp, ideally of at least 100-200W in power handling, the midrange and HF unit are never likely to see more than a watt in typical use.

You can further lower the workload on the MF+HF amplifier by taking advantage of the fact that there tends to be more musical energy at the low frequencies than the high.  By moving the high pass filter from the speaker crossover to before the amplifier input, you can reduce the load by 3dB or more, depending on the music.  Do note that this is fine for a midrange like the TD15M Apollo which has insane (>500W) power handling as even if the amplifier goes DC, the speaker won't care, but it's a really bad idea to DC couple any kind of high frequency driver without suitable protection in place.  Be warned that it can be an expensive lesson...

The high power demands of the bass driver realistically limit choices to a Class AB or a Class D solid state amplifier, which aren't very interesting from a purist point of view, and this drive unit is only covering a few hundred Hz with fairly quick rolloff, so let's not concern ourselves with that for now.  What is interesting is the watt for the MF+HF, which covers roughly 250Hz onwards... the bulk of the sonic spectrum.

So we want a good watt... how to get it?  The immediate answer is Class A operation... whether it be a single output device or a push pull pair, "always on" operation yields the lowest distortion, but unfortunately also the lowest efficiency.  For a single clean watt, we can sacrifice efficiency and still keep power consumption within manageable limits.

(Do note that I've seen a Class D amplifier that idles at over 20 watts, so it can be dangerous to make assumptions purely based on topology!)

There are some very well known Class A designs that put out a few nice watts, and I've been endeavouring to build them up to try with the speakers... time to build, listen and measure... in that order.  :)

Speaking plainly...

Audio is difficult.  This is the conclusion I've managed to come to after a decade or so of practising in the art.  Many systems are good at one or two things, but to cover all the bases is incredibly hard.  Being pragmatic is to decide what particular aspects are most important to you, and to focus on achieving those primarily.

Take my speakers, for example.  I'm certainly not a speaker designer by trade, but I know what's most important to me... low distortion and an even tonality with no undue emphasis, particularly in the upper midrange.  I also hate cabinet colouration, which led me to come up with a speaker with no cabinet, or baffle for that matter.  Throwing away these things greatly hurts efficiency and creates potential nulls in bass response, but that is the price that needs to be paid for a small, relatively light footprint which doesn't suffer from the usual smearing of sound as the cabinet resonates.

Losing efficiency in one area of the design like this requires high efficiency, high power handling units to compensate for the shortfall.  I've long since used Lambda Acoustics TD15M Apollo for mid/bass duty (a big 15" unit with stupendous power handling and a curvilinear cone for wide frequency response) but the thought occurred to me that it could be used for a pure midrange.  With the Raal ribbons on top, that is potentially a 95dB/W midrange upwards block, which could be driven by its own small amplifier.  No baffle step compensation should be necessary here, as that can be incorporated into the bass driver crossover, which will have to be driven relatively hard to achieve any bass.

I had been mulling over the idea of a baffle-less speaker for a while, but this created a problem... how to support the drive units?  A bit of a discussion was had with a talented carpenter called Russell who's turned some of my (rather poorly drawn) ideas into reality before, and eventually settled on the idea of a V shape as a support.  This had a potentially interesting property of dividing the rear wave energy, which may help ameliorate nulling problems to a degree.  He builds out of thick birch plywood, which is an excellent material acoustically - a light year away from the typical MDF used in most speakers.

The magnets of the drive units neatly slot into the supporting holes.  The idea was to bolt the drive units together, but in an attempt to reduce transferral of energy through the front baskets, oil-based clay was used to cement the units into the mounting holes - this proved sturdy enough to support the drivers, with some sorbothane spacers keeping things at the right height.

I haven't said much about the 15" bass drivers so far, which are high Q units similar in design to the TD15Ms... I had a bit of a trial with them, discovering that one unit had a cracked basket long after I'd purchased them - the speaker company in question was not particularly helpful in the matter, but thankfully I was able to epoxy the basket to what appears to be adequate strength, and distortion performance appears unaffected.

As said, the nature of the design means that the bass drivers need to be heavily equalised in order to produce any reasonable bass.  In the case of these drivers, a low pass filter was pretty much all that was needed to bring things into line, after attenuating the midrange+HF suitably.

The attenuation of midrange+HF to get a decent bass response worked out to approximately 12dB, which really isn't too bad, although does mean the effective efficiency of the bass driver has been reduced to roughly 78dB/W!  With roughly 200W power handling, that is enough to play reasonably loud, but it won't blow the house down... that was not on the requirements list.  :)

The crossover is relatively simple... second order on the bass driver, first (high pass only) on the midrange, and the tweeter is effectively fourth order, if memory serves.  Response looks something like this, about 15-20 degrees off axis...

It's a fairly even response with mainly dips rather than peaks, and a gentle roll off at the top end unless you're sitting bang on axis... the limited vertical dispersion of the ribbon is a bit of a price to pay for the excellent sound quality it provides, though this could be ameliorated with an "ambience tweeter"... something I've been meaning to try for over a year now but haven't got around to!

Once the crossover was gotten into a reasonable shape, I stopped messing with them and just started to listen to music... which is how it should be, really!  They have their limitations (primarily the small sweet spot), but perform well with a wide range of musical genres.

The big gap in efficiency between bass and midrange+HF suggests biamping would be an excellent idea - perhaps a high efficiency Class D driving the bass, and a high quality small Class A driving the midrange+HF.  Now there's a thought...