Yet Another LME49811 + STD03 Build

Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.
Member
Joined 2009
Paid Member
Folks,

I have previously tried the LME49810 with 2SC3281, 2SA1216 output devices, MJE340/350 drivers and failed, I figured I should try the LME49811. Well... "failed" is a bit of an overstatement. I had some hum issues due to poor layout and while debugging that, I came across the LME49811 and the Sanken STD03N/P ThermalTrak devices. As I also experienced questionable bias stability in my LME49810 circuit, I figured I would try the ThermalTrak devices.

Anyhow... The goal here is to build an amp that will drive my subwoofer (8 ohm, Dayton RSS315HF-8), though, the amp will be built as a stereo amp so it can be used in other applications if need be. The amp will have differential input (XLR). I will be reusing the chassis, input board, and power supply of a Parasound A23 that a friend of mine gave me after his ex-girlfriend broke it by spilling water in it by accident. So the power supply is fixed at +/-60 V, 1 kW. I'll be reusing the original heat sinks -- which are surprisingly small. Nice extruded aluminum, but not anodized. I'm guessing their thermal resistance is on the order of 0.7~0.8 K/W. The heat sinks have thermal switches on them, that judging from the part number would switch off at 95 deg C (damn hot!). The original amp specs claim 125 W, 8 ohm and reproducing that in my own design would be fine & dandy.

I managed to get my paws on a small surplus of the STD03N and STD03P. I have enough that I could go with three pairs per channel, but I don't see a point of using more devices than necessary. Using the textbook equation, Pdiss = 2*Vcc^2/(pi^2*RL), I calculate the power dissipated in the output stage to be 91 W for 8 ohm load, 182 W (!) for 4 ohm load. Granted, with 4-ohm load it'll also deliver some 300+ W to the load, but still... 182 W dissipated!! :eek: So I'm thinking to use the temperature sensor to make sure the output doesn't fry. According to the STD03N/P data sheet, each transistor can dissipate about 60~65 W at 100 deg C (where I take the thermal switch will have tripped). So it looks like two pairs should be enough. Is there an advantage of using three pairs? Does anyone (Panson_hk maybe?) have some data to show? I recall Douglas Self muttering in his Power Amps Design Handbook that there wasn't much of an advantage of going beyond two pairs...

Another thing is how to handle multiple ThermalTrak devices. I know Sanken advises that each pair has its own bias pot. Fine! But the recommended operating point of 2.5 mA per diode stack would suck up 5 mA of the 6.5 mA the LME49811 can deliver, leaving, only 1.5 mA for the output devices. Although, I have the 'Y' version of the STD03's (Hfe > 8000), I still find it wasteful to burn more current in the bias circuit than I'm delivering to the output stage. And using an LME49810 just to drive a stinkin' diode stack seems even more wasteful. Hence, I'm thinking to use just one diode string for the two/three output device pairs. I'll use the diodes in the "inner" devices - i.e. the devices closest to center of the heat sink as I assume they will be the hottest. Does this sound like a reasonable approach?
Or should I just use the two diode strings in parallel and specify the cap across the diode stack to be large enough that the bias voltage doesn't sag significantly on large transients and quit worrying about it?

Speaking of bias. I calculated the resistor between the SOURCE and SINK pins as follows. I measured the bias current sourced by the LME49811 by attaching an ammeter between SOURCE and SINK. I measured the bias current to be 7.9 mA. I wanted 2.5 mA in the ThermalTrak diodes as Sanken says this is the recommended operating current. Figuring the 200 ohm bias tweak pot would be centered, I calculated the expected voltage drop across the diode stack and the pot to be 2.495 V. Thus, R = 2.495/(7.9E-3 - 2.5E-3) = 462 ohm. 470 was used. What I don't like about this, tough, is that the current through the resistor and diode stack will vary depending on the base current drawn by the output stage. Would it matter much if I ran the full LME49811 bias current through the diodes? I could scale up the current in the output devices by the same factor... Or is there a better way of doing this?

BTW: I notice that many people use a small (10 ohm seems to be typical) resistor in series with the base of the output devices. What's the purpose of this? Taming parasitic oscillation?

I have built a small proof-of-concept prototype (see attached). So far I'm impressed. Once I had the math for the bias circuit down, it was really a breeze to get going. I don't hear any hiss or hum in the speaker -- something that's confirmed by a quick look with a spectrum analyzer. The distortion is 0.002~0.003 % across most of the frequency spectrum at 1 W, 8 ohm output. Note that this is basically the noise floor of my HP8903A distortion analyzer. This is with a setup that's wired up on a solderless breadboard held together by double-stick tape and bailing wire. I did wrap it in grounded tinfoil for the noise measurement, though. That knocked the little bit of induced 60 Hz hum down from -93 dBV to below the noise floor. Nice!!!

I have included the main components on the schematic. In addition, I'm using 29 pF (22||7 pF) for the compensation cap, 100 uF supply bypass with 10 uF || 100 nF at the chip. The final version will include the 75 pF + 3k3 noise gain compensation network that Audioman54 mentioned elsewhere. And of course, I'll have more output devices, DC servo, etc., etc.

The distortion plot is 1 W into 8 ohms with +/-40 V supplies (HP 6228B lab supply), 80 kHz bandwidth. Note measurement limit is about 0.002 % due to the noise floor of the HP 8903A.

Anyway. I'm looking forward to reading your comments on number of output devices and bias scheme. Meanwhile, stay tuned as the build progresses.

Thanks,

~Tom
 

Attachments

  • LME49811amp_Prototype_TC20100214.jpg
    LME49811amp_Prototype_TC20100214.jpg
    128.6 KB · Views: 5,297
  • LME49811amp_Schematic_Prototype_TC20100214.jpg
    LME49811amp_Schematic_Prototype_TC20100214.jpg
    103.3 KB · Views: 6,053
  • LME49811amp_Distortion_80kHz_1W_TC20100214.png
    LME49811amp_Distortion_80kHz_1W_TC20100214.png
    29.7 KB · Views: 4,930
  • LME49811amp_Noise_1kHz_TC20100214.gif
    LME49811amp_Noise_1kHz_TC20100214.gif
    9.5 KB · Views: 4,671
  • LME49811amp_Noise_100kHz_TC20100214.gif
    LME49811amp_Noise_100kHz_TC20100214.gif
    9.3 KB · Views: 4,630
this is just a suggestion.
Miss out the 470r initially.
Add a 10r in series with the bias voltage adjustment pot.
Measure the current through the diodes by monitoring the 10r voltage drop.
Bring in the new value of the bypass resistor to reduce the diode current down to near Sanken recommendation.
Check the temperature compensation change from open circuit 470r to the new value 470r.
 
Dear Tom,

Here an example of my working STD03 amplifier design. To obtain more power I use a bridge configuration. Also for driving a subwoofer. This design works still flawless. I am stubborn and tied the diode strings from both pairs together. I stole this from Arcam who does exact the same in their P1 amplifier with two pairs of STD03's.

Ps. The STD49811 can't deliver enough current to fully drive two pairs on STD03's. Better go with the LME49810.

With kind regards,
Bas.

Ps. DC servo not included since this amplifier is used in an active speaker system, where DC is removed in the crossover/input stage ;)
 

Attachments

  • STD03.jpg
    STD03.jpg
    87.5 KB · Views: 3,944
Member
Joined 2009
Paid Member
Bas! I greatly appreciate your input.

Here an example of my working STD03 amplifier design. To obtain more power I use a bridge configuration. Also for driving a subwoofer.

That's exactly what I'll be doing. As you've figured out in your design also, having balanced inputs makes it real easy to bridge a stereo amp as all you have to do is swap two wires on the input...

I am stubborn and tied the diode strings from both pairs together. I stole this from Arcam who does exact the same in their P1 amplifier with two pairs of STD03's.

I can understand why Arcam didn't want to use two separate bias pots as it doubles the cost associated with the bias current adjust. But for my little one-off hobby design, I'll go with two pots if I'm using two diode strings. The current through the diode strings will be the same regardless of the number of pots.

Ps. The STD49811 can't deliver enough current to fully drive two pairs on STD03's. Better go with the LME49810.

Curious :blackcat: here: What do you base that on?

Here's how I think about it: I've attached the Hfe curve for the STD03. The worst case (-30 C, 10 mA Ic) Hfe is about 80. 0.01/80 = 125 uA base current. Surely, the LME49811 can deliver 125 uA... In the other end of the spectrum... Max Ic is 15 A per device. So 30 A for two pairs. 30/8000 = 3.75 mA. If I run two diode strings at 2.5 mA each, that becomes kinda marginal at worst case output current for the LME49811 (6.5 mA). So a little math to start the day:
One diode string: 2.5 mA. LME49811 w/c current: 6.5 mA --> 4 mA left for the output base current. 0.004*8000 = 32 A. Plenty...
Two diode strings @ 2.5 mA each --> 1.5 mA left for output. 0.0015*8000 = 12 A. That's getting low...

Another way to approach it:
Remember 60 V supply...
4 ohm load: Iout = 60/4 = 15 A (7.5 A per device).
8 ohm load: 60/8 = 7.5 A.
Looking at the Vce(sat) vs Ib curve (also attached), it seems that for 10 A per device I should have 2 mA of base current available per device. This is for an ambient temperature of 25 C. As the transistors will be on a fairly warm heat sink inside a chassis, I'd consider that a reasonable worst case calculation as Hfe increases with temperature. This calculation assumes that the LME49811 can drive the output stage into saturation - which it can't. The actual peak output voltage will probably be a good 10 V down from the rail, hence, easing the output current (and base current) drive requirements a bit.

Sorry for the thesis, but reading through the various LME4981x + STD03 threads, I've noticed *a lot* of chatter about bias current drive and I'm just wondering where all the confusion comes from. Maybe I'm missing something completely fundamental -- in which case please point it out to me.

Ps. DC servo not included since this amplifier is used in an active speaker system, where DC is removed in the crossover/input stage ;)

I'm considering that too. But the DC servo is nice to have in many cases and can avoid the worst turn-off pops. It's not a big deal to add and if I don't like it, I can just pull the servo IC.

Thanks,

~Tom

STD03N_Hfe_vs_IC_TEMP.PNG

STD03N_VceSat_vs_Ib.PNG
 
Last edited:
Member
Joined 2009
Paid Member
Bias measurements

I'm starting to understand why people suggest using the LME49810 to drive the STD03's. If I'm going to use the LME49811, I will be limited to one diode stack. That's alright, though. The resistor that most people suggest to use between the SOURCE and SINK pins concerns me, though, as if it is designed to keep the current flowing in the diode stack at or below 2.5 mA at idle, it will take up all the remaining output current of the LME49811, leaving little to drive the output devices. Basically, the part of the LME49811 output current not drawn by the resistor will have to be shared between the diode stack and the output device base. Hence, I think the resistance between SOURCE and SINK should draw just enough current that the bias can be stabilized but no more. This is probably what leads to the 1~2.2 kOhm seen in most designs.

With no resistor between SOURCE and SINK, the bias is very touchy to adjust. I ended up at about 130 mA (65 mV across 0.5 ohm). With 2.2k it was easier to adjust the bias and I ended up at 100 mA. Slightly better still for 1.5k. I think I'm fairly comfortable with my current setup with 1.5 kOhm between SOURCE and SINK. I did not see any immediate signs of bias stability when tracking the bias current (actually the emitter-to-emitter voltage on the two STD03's) over half an hour. The attached data shows the voltage vs time at 5-second intervals. The step is caused by a change in input voltage to the amp resulting in a 0->15 W step in output power.

~Tom
 

Attachments

  • BiasStability_TC20100215.PNG
    BiasStability_TC20100215.PNG
    39 KB · Views: 2,804
Member
Joined 2009
Paid Member
I think it's time to go full scale... I'll throw together a board that allows for both diode stacks to be used in parallel (with separate pots) and has the room for a resistor across SOURCE and SINK. If I don't want the resistor between SOURCE and SINK, I just don't populate it. If I only want to use one diode stack, I just yank one of the bias adjust pots. Easy... I'll post schematics, results, pictures of exploded parts, etc. as the project progresses.

~Tom
 
Member
Joined 2009
Paid Member
Is con2 2 a cold input from a balanced source or a signal ground?

The input is from a balanced source. I realized after I posted the schematic that I need to bring out the feedback/input ground also.

The 5k6 passes just 8mA to the 411 and the Zener. This is not enough.

Well... The LF411 max supply current (if I recall correctly) is about 3.9~4 mA. Then the output current should be added -- worst case 15V/28kOhm = 0.53 mA. That leaves the zener with about 3.5 mA. Looking at data sheets, it seems the 15V zeners are mostly tested at around 8~9 mA, though, I have yet to find one that actually *specifies* the zener voltage vs current. Anyway... I think you're right. I should probably up the current through the zener some. Go with a 3k3 instead (--> Izener = 9 mA). That would lead to a power dissipation of 614 mW in the resistor. A 2 W is probably OK, though a 3 W type would be nice.

How much interaction is there between the two bias strings and their adjustment pots?

That's a good question. It has crossed my mind as well. I'm really just going with what Sanken recommends at this point, thinking that if I don't like it I can pull one pot and run on one diode string, connect the two strings in parallel with just one pot, etc., etc. The changes would be easy to make with a little "blue wire" and shouldn't hurt circuit performance in the end. I am quite curious to see how the bias adjustment turns out. I fear it'll be a ping-pong game between the two pots, but we'll see...

I may tweak the component values for the RCs on the supply to the LME49811 due to board space constraints. But I haven't decided that yet. Unfortunately, I already have a chassis made for this project so I'm a bit boxed in as far as board real estate and power transistor placement go.

~Tom
 
So... Off the solder-less bread board and onto a PCB. I skipped the local supply filtering (R2, C3, R22, C13) due to space constraints. I'm currently running off of +/-50 V (the max my lab supply can provide). I'm seeing 0.007 % THD+N at 35 W output power, 1 kHz, 8 ohm load.

I fiddled with the bias adjust. With the resistor across the diode stack - R12 in the .pdf schematic - stuffed with 2.2 kOhm, the bias is very, very hard to get equal between the two pairs of output devices. I tried placing the two pairs of diodes in parallel with just one adjustment pot and that worked a bit better. The emitter-emitter voltage (across 0.44 ohm) was about 10 mV different between the two pairs of output devices. So 30 mV for one pair, and 40 mV for the other. Desoldering R12 and leaving it open, while having two adjustment pots (basically as drawn in the schematic less R12) made it possible to get the two emitter-emitter voltages to match within 3 mV. So I think that's what I'm going with. It will result in almost 4 mA through each diode stack, but the bias stability is NICE and the response time to a step in output power seems instant.

Attached is the emitter-emitter voltage response to a 35 W step in output power. The idle current is about 100 mA per pair of output devices (40 mV across 0.44 ohm).

~Tom
 

Attachments

  • LME49811_Amp_BiasStability_TC20100307.PNG
    LME49811_Amp_BiasStability_TC20100307.PNG
    19.6 KB · Views: 488
The emitter-emitter voltage (across 0.44 ohm) was about 10 mV different between the two pairs of output devices. So 30 mV for one pair, and 40 mV for the other.
says that one pair is passing 33% more current than the other pair.

The parallel output devices must be matched for Vbe at the operational current.

This requires both devices to be fed with the same Vbe and for the currents (Id) to match. I set a tolerance of <5%.
It does no good to set the current and try to measure the Vbe.
It does no good to try to measure hFE.

Us cash strapped amateurs must match DUT to REF to get matching sets.
 
Desoldering R12 and leaving it open, while having two adjustment pots (basically as drawn in the schematic less R12) made it possible to get the two emitter-emitter voltages to match within 3 mV. So I think that's what I'm going with. It will result in almost 4 mA through each diode stack, but the bias stability is NICE and the response time to a step in output power seems instant.

Attached is the emitter-emitter voltage response to a 35 W step in output power. The idle current is about 100 mA per pair of output devices (40 mV across 0.44 ohm).

~Tom

In my case for 49811 with two-pair ThermalTrak, I recall that there was about 2 mV difference for 22 mV across 0.22R emitter resistor.

The step response looks good!

I have evaluated output stage with four-pair ThermalTrak. It offers lower distortion for heavy loads, say, high power into 4 Ohm.
 
says that one pair is passing 33% more current than the other pair.

The parallel output devices must be matched for Vbe at the operational current.

Us cash strapped amateurs must match DUT to REF to get matching sets.

True. Using matched devices would allow for better current sharing. Not only do the Vbe's need to be matched, the Vf of the diodes also need to be matched. Though, I would expect Vf to track Vbe fairly tightly -- that's the whole point of this ThermalTrak thingy. Being a cash-strapped amateur myself -- at least regarding this project -- I don't have a large enough pool of devices to match from. Hence, my desire to follow Sanken's recommendations of separate bias adjustment pots for each pair. I think if one goes through the trouble of matching the output devices, one should also match the emitter resistors -- or at least get resistors with a tighter tolerance than the +/-5 % I have...

In my case for 49811 with two-pair ThermalTrak, I recall that there was about 2 mV difference for 22 mV across 0.22R emitter resistor.

So a 10 % difference. Thanks! That's a good data point to have. I'm fine with a 10 % difference in device current as long as none of the two bias chains exhibit any tendency to run away thermally.

The step response looks good!

I was really quite impressed by it! On my previous setup (LME49810 + MJE340/350 driver + standard Vbe multiplier) it took over an hour for the bias current to decay after a load step. With the LME49811 + STD03, it's nearly instant. The slow settling following the step is basically the thermal time constant of the heat sink.

I'm looking forward to getting the amp on the final power supply (+/-60 V, 1 kVA) and running higher powers.

I have evaluated output stage with four-pair ThermalTrak. It offers lower distortion for heavy loads, say, high power into 4 Ohm.

Nice. I was also impressed by the low-power distortion measurements I took. I measured the output spectrum at 100 uW output power and the harmonics were below the noise floor on my analyzer (85-ish dB dynamic range). The LM3886 by comparison shows quite strong 2nd and 3rd harmonics under the same conditions.

~Tom
 
I was really quite impressed by it! On my previous setup (LME49810 + MJE340/350 driver + standard Vbe multiplier) it took over an hour for the bias current to decay after a load step. With the LME49811 + STD03, it's nearly instant. The slow settling following the step is basically the thermal time constant of the heat sink.

~Tom

The driver temperature not being tracked causes some issue. I am studying this effect.
 
Why not parallel the two halves of an LM4702?

To my knowledge, the LM4702 is basically two LME49811's in one package. I suppose one could connect two in parallel to get higher output current, but I don't really need that. My LME49811 delivers about 8 mA (worst case according to the data sheet is 6.5 mA). So with a worst-case Hfe of the STD03Y that I have, I should have minimum (6.5 mA)*8000 = 52 A output current. Yeah, Hfe depends on collector current (yadda, yadda), but looking at the Hfe curve, I still don't think the amp will be current limited. 52 A is more than the two devices can handle anyway...

The other possible interpretation of your statement, is that you suggest building two power amps using one LM4702. Each power amp would have one pair of STD03's in it. The two amps would then be connected in parallel. This would alleviate separate bias for the two output device pairs, hence, their bias currents could be dialed in to be dead on.

However, with amps in parallel, it is a challenge to ensure that the multiple amp channels share the current evenly. It's not an impossible challenge to solve, but getting accurate load sharing between multiple amps is probably as difficult as getting accurate load sharing between multiple output devices in one amp. I like the simplicity of the LME49811 plus a handful of external parts, so that's the battle I've chosen.

~Tom
 
the RF filter is set very high. F-3dB~16MHz.
Is there DC blocking in the source component?
C2 (DC block on NFB is acting as a high pass filter) seems a bit small. try 3 to 5times bigger. C2 should be connected to the input/signal ground.
R8 (2k2) may need to be adjustable to tune in the temperature compensation.
How does the inverting half source current for the +IN & -IN pins?
What is the resulting output offset?
I don't understand how the inverting side is set to unity gain. Is it?
 
Status
This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.