2nd Spring -- another stab at the SE choke-fed Class A Power Amp

Status
Not open for further replies.
We recently had a blizzard (by Portland standards, that is, the city has no equipment to handle heavy snowfall). So I spent a week confined to places we could walk to, and spent most of it with cabin fever, playing on the internet. I ran across a new website for circuit board layout called PCBWeb.com and tried it out. For a test, I took a quick stab at a revisit of the "Spring Amplifier", a nice sounding solid state class A power amp I came up with a while ago.

This is a topology I've been obsessed with for some reason a while now. There was an article about it in Jan's "Linear Audio" magazine last year. I've since done some enhancement on the original (built largely on perfboard) and have been listening to a couple of 55W mono builds of the design for a while now.

Anyway, in about a day a schematic was entered and a board layed out -- pretty nice layout package, the pcbweb is -- I like it! The user interface can be a bit infuriating at first , but pretty easy once you figure out its oddities. I'd also had read some good things here about the pcbway.com board house, and so browsed over there out of curiousity to see what it costs to have boards like this etched. Which turns out to be a LOT less expensive than I expected, so of course before I had even given it much thought, I'd hit the button to order 10 of the main boards. Along with 10 output capacitor bank/clipping detector boards for them that I added soon after at only about a dollar each. So, without planning to, I had a project going. And it's been fired up now, so -- here's documentation about it.

Comments, corrections, suggestions, questions, wisecracks welcome!

(I notice that some of the images in the thread are a little too small to easily read. I don't know how to change viewed image sizes, and it's too late to edit most of this string of posts, anyway. But if you're curious to see more detail, you can look at them at the Gallery - http://www.diyaudio.com/forums/gallery/showgallery.php/cat/2243

2nd_Spring_fig1.png
The layout started as a basic version of the enhanced Spring amp, trying for a kind of minimal, single output transistor, relatively low power setup. But once the board layout got very far along, I thought to shift to a more flexible layout, allowing for some of the enhancements I'd come up with (which could be worked around if more simplicity was still desired). So support was added for paralleled output sections for higher power, buffering for better bandwidth and linearity, a compound output for higher damping factor and lower distorton, a fancier input stage that made for still better overall performance, and a time-stretched clipping detector.

The basic Spring Amplifier is a single voltage gain stage, which drives the speaker via a unity-gain follower like this:
2nd_Spring_fig2.png
The circuit all kind of hangs from the end of a feed choke or inductor, hence the name "The Spring" (pretty clever, huh? Yeah, not really I know, but it's the best name I could think of). All the audio components increase or all decrease conduction in the same direction, and the waveform is linear at every stage. If SE class A is the goal, the topology allows for respectable performance: good efficiency from the inductor feed, high damping factor from the follower output, a low but triode-like distortion profile (dominant 2nd order, with lower third and then others quite far down), no audible noise at all, output DC protected. And all using only a single high-current power supply and operating without a feedback loop. A detailed discussion about the topology can be found in the Linear Audio article. It's also a fun and inexpensive amp to build.
 
Last edited:
For any practical build of this, a way is needed to keep the output devices' idle power consumption under control, so that is watched by an integrator circuit (not shown above) which gradually steers the current source (is shown above) which controls that. The single-ended class A topology operates with constant power consumption -- what doesn't go out to the speaker gets eaten up by the output devices. Also, the output device shown above (as a PNP emitter follower for simplicity) really would like to have a beefier feed than the cascode stage alone can supply, so it would be best configured as a Darlington or as a Sziklai stage instead. I went with a Sziklai. And distortion was found to improve further by first buffering the voltage signal through a FET follower as well.

The input stage can also be configured as a Sziklai using an added PNP transistor (well worth the added 50 cents or so). And the cascode NPN transistor can be made a Darlington pair, which assures all significant signal current from the input stage is linearly transferred to the output stage -- the result does show up as improved distortion profile, and again, costs almost nothing to add.

A further change since the Spring article in Linear Audio, I separated the power supplies of the output follower stage(s) and of the input voltage stage. Power efficiency directly translates to power supply, heatskink, and chassis sizes - which coincindently also typically account for most of the expense of a power amplifier! Better power efficiency of the output stage can be had when the voltage stage can pull its drive all the way down to the output common. So the input stage sits at a somewhat lower voltage, operating off a small, very-low current, supply.

You'll also notice some other adornments in the real circuit -- grid stopper resistors on FETs, some diodes and capacitors to control the power-up behavior, capacitors to control rolloff, and clamping parts to protect sensitive circuits (utilizing the EB junction of NPN transistors as inexpensive low-leakage zeners). LEDs are also used to generate some voltage offsets, because they're convenient and can allow the same circuit to operate from a wide range of supply voltages for different output power capabilities.
2nd_Spring_fig3.png
Q2, Q3, and Q7 are the NPN transistors doing duty as clamps. Q4 (optional, if removed short R5) makes Q1 be the control of a Sziklai pair. Q6 makes the cascode transistor a Darlington. Either blue or white LEDs can be used for low cost voltage references (want something with around a 3V drop).

Q8 is a current source, programmed by the voltage kept across C2 by the servo integrator. A low current supply (in the range of 11 to 20V) connects at "J3" (which is just some solder pads with matching strain-relief holes, like the other J-numbered connectors).

A single board supports a pair of output devices (each part of a Sziklai follower); addional boards can be used to add further pairs of output followers or they can be hand wired (they're only two transistors and two resistors each). The KSA1381 transistors (Q10 and Q12) should NOT be mounted on their matching NPNs, keep them off some to minimize any tendencies toward current hogging. The schematic, by the way, is of a revised layout (that I didn't get fabbed at least yet), so the reference designation may not match the picture in the first post here.

The output capacitor (and a stabilizing RC) can be remote mounted or can be done with the following Output/Clipping Detector board.
 
Output/Clipping Detector board-
2nd_Spring_fig4.png
In SPICE sims and in the unit built, L1 hasn't been needed so is just a short piece of wire. It's in the layout 'just in case'. C7 and R14 might also be optional, but SPICE shows some peaking with some capacitive loads without it.

The clip detector senses even very brief output clipping by monitoring the output transistor's voltage and current. When either runs out for some microseconds, capacitor C5 discharges and lights an indicator LED (connected to J4) for about a half second afterwards so you can see that it happened -- assuming you want to know that it did. It's there for neurotic listeners like me who tend to wonder when using a limited output SE amp whether they might be clipping just a little. The LED staying dark means they're not. Though when the amp is clipped even moderately that seems hard to hear. So maybe better to NOT use the indicator? Another option.

Being a Single Ended power amplifier, a Spring amp will usually clip asymmetrically. We don't get a linear transition to class AB and higher power like push-pull type class A amps might give under lower impedance loading. The asymmetry depends on the load impedance. With higher than designed-for load impedance, the voltage on the output (capacitor side) can swing more positive than the supply's magnitude, but it's peak can't ever swing more negative than the nominal. With lower than designed-for load impedance, the output current's negative peak can be stronger the amp's idle current, but the positive current peak can't ever be more than the idle current. So when clipping with real speaker loads, the output power envelope usually can continue to vary, though with some distortion, past the clip point. Maybe that helps hide audible effects of clipping until it gets more egregious, or maybe it has to do with the less sudden onset of distortion (compared to higher feedback amplifiers) near clipping.
 
The parts that cost money (though not too bad in this design) --

The amp needs heat sinks good for at least 2 1/4 times the maximum unclipped output power (and more like 3 1/4 times, if making lower power or lower impedance targeted designs) . I've been using some inexpensive ("hobby-justifiable") amplifier chassis -- google "2412B Full Aluminum Enclosure" at aliexpress. They include pretty good heat sinks, are well made, and look good. Shipping is about as much as the chassis costs, though! The nice big-heatsinked chassis at the diyAudio store are worth considering, particularly if you are going for higher power (the 2412B is good for a stereo 18W/ch amplifier, or a mono amp of 50W+ but running pretty hot).
2nd_Spring_fig5.jpg
2nd_Spring_fig6.png
The power level also determines the size (and cost) of the Spring inductors (chokes) needed. Things to watch for are the inductance (more is better, need at least 25mH) and the max unsaturated currrent rating. These are going to be relatively big gapped-core inductors. I started laying out the board planning on the inexpensive ($14 at the moment) Triad C-56U inductors, 35mH at 2A. By pushing the current a little (the inductance seems to still be dropping only smoothly even at aout 2.5A at least for the ones I have) I get about 16W into 8 ohms out of these. Two channels using those will fit in the 2412B chassis -barely!- along with a small switching power supply -- google "120W 24V 5A Little Switching Power Supply", at all of about $12 shipped.

By the way, a high frequency switching type supply is a very good choice for Spring type amps. They're silly inexpensive (look for ones made for LED lighting), regulated, somewhat voltage adjustable, have very little low frequency ripple, and in this amp there's a mondo big inductor that sits between the supply and the audio signal path-- high frequency ripple is just NOT going to get through. I could of course have used a big "linear" (in other words, 60Hz switching!) supply instead if I had either the parts for it or the money waiting to be used. With an unregulated "linear" supply, the filter caps should be substantial though, because the spring inductor won't prove a lot of ripple isolation at power line frequencies. The lower current supply (good for 30mA or so) could also be either "linear" or a small switching module, since the power supply rejection ratio of the amp is pretty good for that over the audio range. I used a switcher, but followed by a three-terminal linear regulator 'just in case'.... the neurotic listener thing, again.

Some inductors that could work with these amplifiers are listed below. Lower resistance ones help to improve efficiency a little (but don't much affect heatsinking). With higher load impedances, particularly if operating into low bass, higher inductance values are advised. There's a vendor at eBay at the moment selling off some used Signal CH-4 and CH-6 inductors at good prices.
2nd_Spring_fig7.png

Going to higher power affects heatsinking, power supply voltage and current capability, the number of output devices and the inductors. Here's a table estimating supply and output set (and heatsink) needs. I haven't done more than about 60W so far in a Spring amp, but I expect versions to 100W, maybe more, should be doable. But with voltages over 36V, a higher voltage part for the cascode might be advised.
2nd_Spring_fig8.png
 
Here are some measurements made for a version configured for operation (where the clip light just goes on) to around 15W -
2nd_Spring_fig9.png
Frequency Response:
2nd_spring_fig10.PNG
Harmonic Distortion vs. Power level:

An assumed benefit of SE class A amplifiers is that the distortion continues to drop as power level drops, and that the distortion products are primarily low 2nd and 3rd order. So it seemed worthwhile to try to check that and to see what the effects of "large inductor" vs. "smaller inductor" might be. So I did that for both the Triad C56-U (35mH, 2 Amp) and the Signal C-4 (70mH, 4 Amp ). For both cases, the operating idle current was 2.2ADC.

I'm measuring using a USB sound card (TASCAM U2-2) based FFT, which is supposed to run at 24bit -- but apparently the Windows 10 MME driver for it only works at 16bits 🙁 . Also, distortion products (particularly 3rd and 5th order) tended to get a erratic at lower levels, which will be in part due to resolution and partly to distortion generated by the analog audio stages of the equipment. Some of the curvatures at low levels probably indicate some cancellation or summing with those added harmonic generation effects. I made a "twin-T" notch filter and a passive RC lowpass filter for measurement near 1kHz to minimize sound card linearity effects and be able to look lower in level with better confidence, so the 1kHz curves are the ones here that are most likely to be reliable. Too lazy to do that for all test frequencies, though.
2nd_Spring_fig_11.png
All distortion levels are respectably low, though there is something strange going on at low frequencies + lower power levels + odd harmonics using the small sized C56-U choke. Also, the obtainable output power with it at 41Hz was several dB shy because its reactance and operating current is too small to generate full positive output swing.

Here is how the distortion increases when the circuit is driven into clipping. For (my) convenience, both curves are at 8 ohms load and with the CH-4 inductor. The blue curve is with 2.2A bias (so that the clipping occurs with negative input peaks, from running out of current and having a supply voltage a little higher than needed) and the red curve is with the bias increased to 3.26A (so that current clipping doesn't occur and the amp instead clips when it runs out of supply voltage range). The blue and red dashed vertical lines indicate where the clip indicator light fires. With sufficient bias current, the amp can do over 20W/ch with a 23.2V supply (which is as low as the 24V switcher's voltage would easily adjust down to). For around 16W, a slightly lower supply voltage and a little more current would be more optimum. Though, speaker impedances being the varying things they are in real life, I don't get too focused on getting the current and voltage exactly on.
2nd_Spring_fig12.png
 
circuit board files, published at and downloadable from pcbweb (Gerbers can be easily generated from there)--

(These aren't copyrighted, and are free use, but are also supplied with understanding that I assume no liability for what anyone might do with them!)

Revised versions:
2nd Spring Amp board (v1.1) PCBWeb - Second Spring DblSided Flood v1_1

Output and clip detector board (v1.1) PCBWeb - Output and ClipDet Board v1_1

Revised_board.png
revised_clipdet.png


The amp board layout could be made as a single-layer also. There are three obvious jumpers that would need to be added on the top. I flooded the excess area on both sides just because it was easy and free to do, but the copper on the top side isn't really necessary. The output/clip detect board is just single layer. The "busy" sides of the boards with all the traces is the bottom, BTW.

The revisions in the new layout are minor, mostly fixing up holes sizes that are too big for wires and transistor leads. Most important change is to put the four mounting holes of the amp board on isolated flooded planes. On the original, I put them on too-small isolated pads flooded in circuit connections (which should NOT connect to chassis or heatsink) -- too easy for a screw or standoff to break through the solder mask and short out. I also added some more strain-relieved "connectors" to help ease expansion to more output sets and connection to the clip detector. And fixed the polarity (A, K) of the clip indicator LED which was backwards on the v1.0 version.
 
Last edited:
For anyone who might've managed to read this far: I have 3 sets (6 channels) of spare boards of the first version. Version 1.0 boards ar free while they last, one set per request, just PM me here with your address. (The revised version, above, hasn't been sent to the board house, and may not ever get sent. I already have more amps than I'll ever use!).

Here are the files of first version (the one built) boards --

2nd_Spring_fig13.png
 
I revised the connection diagram to make the twisted pair wiring more clear, and changed routing to improve current sharing (when making higher power versions with additional output devises).
2nd_Spring_fig_14.png

Also, here's a bigger pic to see the teeny little distortion component plots a little better --
2nd_Spring_fig_11.png
 
Last edited:
> though looks like it would have high output impedance

I actually had not looked at it in years. Yes, the notional Zout is infinity. Or in buzz-word, "current source". And the input impedance is quite low. And there is a Volt of DC on the speaker.

In fact the bass impedance is 50mH (confirmation for your 35-200mH suggestions) which suggests "slight" damping for typical speaker bass resonances. At 50Hz, 50mH is 16 Ohms which is larger than nominal 10 Ohms but not "Large". DF~~0.6?

AT higher freqs it just rises. A 2N301 may have a collector impedance near 100 Ohms, but this varied a lot. The choke rings and goes capacitive at some high frequency in the broad area of 20KHz. This and speaker-rise may explain the 30uFd "compensation cap".

It mentions "hi-fi". At the time, there would be many "hi-fi" speakers suitable for pentodes with little NFB, so would not be majorly upset by small damping.

A hasty sim with wrong-parts suggests >5%THD mostly 2nd. While tolerable with the right speaker as a very small hi-fi, it may shine as a guitar amp?

Your design, OTOH, looks like point-oh-oh everything and >100 DF.
 
Status
Not open for further replies.