Ovation 250 Power Amplifier
The Ovation 250: A 250W per Channel Audio Power Amplifer (updated 1st June 2013)
Diy Audio is a wonderful activity:- if you cannot afford a $10 000 power amp, you can always build one that comes pretty damn close in terms of sonics for a fraction of the price.
In the intervening years between the early 1970’s and the mid 1990’s, some breakthrough’s in amplifier design were made by a number of researchers and practitioners, and this knowledge entered (and is still entering) the audio design mainstream.
In the mid 1990’s, the first edition of Douglas Self’s ‘Audio Power Amplifier Design Handbook’ was published, and has gone on to become it could be argued, the most important introductory text to audio power amplifier design. It swept aside decades of misinformation and nonsense, placing the ‘art’ firmly in the realm of electrical engineering and physics. Thanks primarily to the work of Self, the 8 distortions that plagued amplifier performance, and therefore had a large impact on their measured performance and sonics, have been codified and their cures laid out clearly and concisely. Furthermore, a greater understanding of the role feedback and compensation design plays in amplifier sonics developed, along with an appreciation of the importance of component selection in achieving the ultimate in sound reproduction.
Execution of the design leading to the final end product is crucial and includes attention to PCB layout and wiring to avoid ground loops and hum; seemingly innocuous things like tapping off the feedback point in the wrong place can result in an 6-8dB increase in distortion, or not following good cable dressing practice leading to magnetically induced noise and distortion.
But audio, like other branches of engineering, also obeys the law of diminishing returns:- if all the rules have been obeyed, then beyond a certain point large sums of money and effort have to be expended to gain only a marginal improvement if any. In well engineered commercial audio power amplifiers, beyond perhaps $3000 or $4000, product aesthetics and ‘sonics’, are the deciding factors – absolute performance specifications differences play very little part.
The Ovation 250 Power Amplifier
The Ovation 250 was my first foray into power amplifier design after a 25 year hiatus from audio. This is a fully balanced topology amplifier that delivers (on 110/220vac) 280W RMs per channel into 8 Ohms, and about 480W RMS into 4 Ohms. The rise time, with the input filter disabled, is 1.5us (10%-90%) in both the +ve and -ve slopes. Its powered by a 2KVA torroidal transformer I had specially wound for the job – its ultra quiet. The power rails are filtered by 47mfd per rail, and the front end is powered off a zener + pass transistor regulated +-80V rail, whilst the output stage runs on +-70V. Being a low feedback design, distortion is not particularly low, and comes in at about 0.1% THD20. I’m not particularly obsessed with ppm distortion – big, fast amps are what I like.
The front panel has a on/off power switch, a clip LED indicator (one per channel) and two status LED’s covering power and ‘system ok’. Around the back, there are two sets of WBT style speaker connectors for each channel (to cater for bi-wiring), a switched IEC power inlet and 2 phono jack type input sockets.
The Ovation 250 weighs in at 38Kg (about 83lbs). There are a few pictures in the gallery.
Ovation 250 Schematics.pdf (updated on June 10 2012)
I originally started looking into this design a few years earlier, and had toyed with single ended to balanced drive and Lin topologies feeding a mosfet (IRFP240/9240) output stage, but eventually settled on a fully balanced (‘FB’) topology using a bipolar output stage based on the MJE21193/4 devices.
Soon after starting this design, I joined diyAudio.com and was introduced to Self ‘s ‘Audio Power Amplifier Design Handbook’ and some very lively discussion around power amplifiers, in the process becoming aware of a lot of new information and techniques. Thus, many of the design aspects are different in some of the designs I am working on now (a 180Watter and another big amp targeting 350W). In those designs, I have retained the FB topology and the bipolar output stage because I am familiar with them and focused on refining the design to improve things like distortion, phase margin and so forth.
My designs feature EF triple output stages and very high slew rates (this is the figure with the front end filter disabled). To achieve this, I run the LTP stage at about 10mA (5mA per side) with lots of degeneration, which of course tends to lower the open loop gain, and thus the loop gain. I run the VAS at about 30mA, and when this is coupled to the high input impedance of the EF Triple, the amplifier easily drives complex, heavy loads. In the Ovation amplifier described here, I also loaded the VAS in order to flatten the open loop gain to beyond 20 KHz. Having said that, this is not a technique I have repeated on the subsequent designs, where I have instead used an inner feedback loop (‘Miller Input Compensation’) around the output pre-driver, VAS and LTP to achieve the same thing in the interests of stability, rather than any specific desire to create a wideband open loop design.
The MJE21193/4’s are very rugged 250W transistors with a large SOA, relatively flat Hfe vs. Ic characteristic (though not in the same league as the MJL3281/1302) and they are cheap. The major problem with these transistors though is their rather pedestrian Ft of 4MHz and this translates into stability issues if you try to run with too much loop gain – in a practical design, you will have to run with about 10dB less loop gain at 20KHz than you would with the much faster MJL3281/1302 devices for similar phase and gain margins. This of course translates directly into higher distortion. Big, fast (high SR), wide bandwidth amps sound good and it’s not going from 0.1% to 0.0001% that makes the difference.
The input is via J7 and into the non-inverting input transistor pair Q18 and Q21 via a 160KHz LPF filter formed by R70 and C19. The input impedance at 1KHz on this amp is about 23k Ohms. The inverting input is formed by Q19 and Q20. All four transistors are degenerated by 150 Ohm emitter resistors (R52-R56). Q16 and Q17 provide the LTP currents which are set at 10mA (so 5mA per side) in this design. This high LTP current, along with heavy degeneration coupled to a low value for Cdom (C29 and C30 at 33pF each), ensures that the input stage can never overload, making this a ‘TIM free’ design. The Ovation 250 was modified in July 2012 for TMC, and this is provided for by the additional 150pF capacitors (C14 and C24) and 1k resistors (R1 and R23).
The balanced output of the LTP pairs are cascoded by Q26, Q27, Q28 and Q31. The front end drive to the VAS stage is developed across R57 and R60 (390 Ohms each). This design does not use mirror loading of the LTP’s, which would raise the loop gain significantly, but require added complexity to overcome the common mode current balance problems (see Edmond Stuart’s write-up’s on this on diyAudio.com). In this design, I returned the inverting LTP outputs to the emitter degeneration resistors (R49 and R51 68 Ohms each) rather than to the rails as is normal practice. This is a trick I saw in a James Bongiorno design and it can offer a few ppm improvement in distortion. The output of the LTP stage feeds a cascoded VAS consisting of Q23 and Q25 in the top +ve half of the VAS and Q22, Q24 –ve bottom half of the VAS. I used legacy BF469/BF470 transistors on this amp which feature very low Cob (critically important for a VAS transistor) and high Vce – I was able to get a whole lot from a work colleague at the time I developed the Ovation 250 in 2007 and early 2007. These transistors are no longer available, and also note that if you try to use them in any LTSpice amplifier simulations, you will get terrible distortion results because the models are not functional. However, in practice, they work just fine.
For the VBE multiplier, also called a ‘Vbe spreader’ (Q15 and Q29 and the associated components), I used a CFP design, because I had concerns that with the relatively high VAS Iq (about 30mA) and possible cross conduction such that at HF, the VAS would ‘open up’ causing serious problems – another reason why the Vbe multiplier is decoupled with a large 33uF (C26). Q15, a BC847C small signal SMD device, is mounted close to one of the output devices collector leads to sense the temperature. I originally thermally coupled this device to the output device collector lead using a blob of heat sink compound, but found that this over compensated the output stage, so as it warmed up, the output stage Iq kept dropping. After I removed the heat sink compound, I got much better results, so that now from cold to very hot, the Iq delta is about 40mA per pair, going from 120mA to 160mA. In my more recent desiogn, the Ovation e-Amp, I used a two point Vbe compensation scheme for better Iq stability over temperature
In the original design, the output of the VAS is heavily loaded with 15k Ohm resistors and I did this after reading an exchange on diyAudio.com about the importance of ‘wide open loop bandwidth’ for good sound. This, coupled with the very heavy front end degeneration, meant that there is a very modest amount of loop gain. Later, I became aware of Robert Cordell’s view about this, and did some further study on the subject and concluded that this in fact was not the case, and as a result removed them, improving the overall loop gain to about 50dB at 1KHz and 25dB at 20KHz, and lowering the distortion by about 12dB. My e-Amp design provides jumper linable loading of the VAS to ground, and this allows me to experiment with loop gain magnitiude simply by inserting or removing jumpers.
The output of the VAS stage feeds a triple emitter follower with Q30 and Q32 operating as class A pre-drivers. I run these at about 10mA by making R65 = 270 Ohms. Originally, I used a speed-up capacitor, C18, across R65, but came to realize that this was not necessary in the pre-driver stage input and removed it. The pre-driver output feeds into the MJE15032/33 (Q13 and Q14) driver stage via 27 Ohm and 1nF RC networks in each driver base. On the original homemade prototype PCB, I could not get rid of the parasitic oscillation in the output stage. After doing some research, I found an application note discussing the oscillation problem on emitter followers and how to cure it. After a bit of experimentation, I settled on the values shown. The -3dB cutoff of these networks is about 5.8MHz, so about an order of magnitude higher than the unity loop gain (ULG) frequency.
The drivers feed into the output stage, and like the pre-drivers, also run in class A and R26 sets the driver stage to about 80mA. This is quite high, and the drivers are therefore co-mounted on the same heat sink as the output transistors Q209 through Q211 (MJL21194) and Q201 through Q205 (MJL21193). In order to tame any parasitic oscillation, each of the output transistors bases has a 3.3Ohm stopper resistor fitted. Although the circuit shows 0.1 Ohm emitter degeneration resistors on each of the output transistors, I actually ended up settling on 0.22 Ohm’s. You need very good thermal compensation and bias stability performance if you are going to settle on 0.1 Ohm output device emitter degeneration devices. Although the Ovation 250 thermal stability is quite good, its not good enough for 0.1 Ohm degeneration resistors. The Iq (adjusted by R2) is set to 120mA per pair (so, 26mV drop across the 0.22 Ohm resistor) after about 15 minutes warm up time.
Basic 2 slope protection is afforded by Q1 and Q7 and the associated components. On my newer designs, I have gon for simple current detection using an opto isolator. I would concede on this design, that the protection is perhaps a little too aggressive.
This design utilizes an output inductor 2uH in the circuit, but later wound for 5uH. With slow output devices like the MJL21193/4, it does not take much capacitive load to pull the output stage pole below the ULG frequency and the consequence is instability or oscillation. The output inductor effectively isolates the capacitive load from the feedback take off point, preventing this problem. However, the low loop gain on this design does mitigate the tendency toward instability with capacitive loads somewhat. Having said that, I really don’t believe a small output inductor is an issue when the speaker cabling inductances can easily reach 4 or 5uH. I originally mounted the Zobel network (R3 and C1) across the speaker terminals, but later wired these directly to the output rail (this is the place where all the emitter degeneration resistors connect together, just before the output inductor. The output of the amplifier is fed to the speakers via 3 16A relays wired in parallel.. I used 3 to ensure low output resistance. On my latest design I have elected to go for a solid state relay based on low Rds(on) 150V mosfets. These are much more reliable than relays when called upon to break high DC currents, as would be the case in a fault condition where, say, one of the output devices failed short to either of the supply rails.
The feedback resistor network values are low, and this is a consequence of running the LTP’s at a high tail current. Although one would expect the input pair base currents to cancel, this does not always happen in practice. I found about 30mV of offset on the output (input LTP transistors not selected), and R31 and associated components (not shown on the circuit) simply allows this to be easily dialed out. I’ve measured the output offset from cold (about 25C) to hot (circa 55C) on two occasions over the last 2 years, and the offset remains below 5mV. No need for servo’s here, and I am confident, with careful layout, this design could be direct coupled and still show very little drift and offset over time once dialed out.
For the control board, I used an NXP 89LPC922 8 bit microcontroller. This connects to each amplifier board via a 20 way ribbon cable, and also provides the high voltage supply for the front end. The controller looks after inrush current limiting (powering up a 2KV transformer without in rush protection trips the mains CB every time), DC offset detection and protection, output stage clipping indicator (this is also handled by the uController) and speaker relay delay’s.
Well, How Does it Sound?
I was in Yodabashi (6 floors of consumer electronics here in Tokyo with some great sound rooms and a big selection of audio) a few months ago and heard a fantastic high end Denon mosfet amp with a big pair of Tannoy speakers. I have never associated Tannoy’s with great sound (apologies to Tannoy fans out there – I have just never heard any prior to this occasion). The imaging was absolutely outstanding (holographic) and top end was smooth as silk, and I thought ‘Wow, that’s absolutely wonderful’. Later that day, I put my system on, dropped a CD in and listened. ‘Wow’, I thought, ‘that’s wonderful too’. It sounded different, but it sounded great. For me, as an audio designer building for myself, that’s what counts – I can have fun designing and building audio gear, and it sounds as good as some great commercial equipment out there.