Hello Chris
One thing that is missed upon about achieving low THD is that you by default have to have excellent physical execution of the design, your cannot achieve low THD with a poor PCB layout and low noise transformers etc ,etc.
Is this what you mean are or are you talking about components you use to achieve sound quality like capacitor and resistors.
Regards
Arthur
Hi Arthur,
I mostly mean the physical layout and a power supply that doesn't generate excessive HF noise. Wire dress is critical, as you've pointed out, as well as the PCB trace thicknesses and layout. PCBs that use some jumpers may perform much better than layouts that avoided using jumpers at all costs. Vias are very suspect in some cases as well.
On parts, there are some we know enough to avoid these days. Part construction and material is far more critical than the name on the outer covering for instance. Almost all film caps have metalized ends these days, excepting some "N.O.S." beauties. Therefore, they all have lower inductance. Metal foil should be better than metalized film types when there are high currents to deal with, common sense things like that.
I know that some designers must bow to whatever the market demands in order to keep the orders rolling in. There may be a difference between what they will say publicly, and to another person in the same field in private. That entire situation is what makes audio a circus with quickly changing fads, and even contradictory "truths" from time to time. The truth is that many older products can be updated, and they sound every bit as good, or better, than the current crop of magazine favorites.
Anyway, the state of the industry was created by that same industry over a period from the mid 70's 'till today. It shows no signs of understanding how much clientèle they have blown off as yet. Older buyers are tired of the games and newer buyers are opting for convenience and low prices. Let's see where things end up, I already lost all bets as I felt the audio industry would crash and burn before now. Just try and find a good technician these days.
-Chris
Edit:
I should mention that some designers are producing rather better products these days. The industry has a better understanding of what they ought not to do, as well as things that improve performance in a real and measurable way. We're very lucky to have some of those people here, willing to post. They are sharing information that has not really affected much in the mass market as yet.
I mostly mean the physical layout and a power supply that doesn't generate excessive HF noise. Wire dress is critical, as you've pointed out, as well as the PCB trace thicknesses and layout. PCBs that use some jumpers may perform much better than layouts that avoided using jumpers at all costs. Vias are very suspect in some cases as well.
On parts, there are some we know enough to avoid these days. Part construction and material is far more critical than the name on the outer covering for instance. Almost all film caps have metalized ends these days, excepting some "N.O.S." beauties. Therefore, they all have lower inductance. Metal foil should be better than metalized film types when there are high currents to deal with, common sense things like that.
I know that some designers must bow to whatever the market demands in order to keep the orders rolling in. There may be a difference between what they will say publicly, and to another person in the same field in private. That entire situation is what makes audio a circus with quickly changing fads, and even contradictory "truths" from time to time. The truth is that many older products can be updated, and they sound every bit as good, or better, than the current crop of magazine favorites.
Anyway, the state of the industry was created by that same industry over a period from the mid 70's 'till today. It shows no signs of understanding how much clientèle they have blown off as yet. Older buyers are tired of the games and newer buyers are opting for convenience and low prices. Let's see where things end up, I already lost all bets as I felt the audio industry would crash and burn before now. Just try and find a good technician these days.
-Chris
Edit:
I should mention that some designers are producing rather better products these days. The industry has a better understanding of what they ought not to do, as well as things that improve performance in a real and measurable way. We're very lucky to have some of those people here, willing to post. They are sharing information that has not really affected much in the mass market as yet.
Last edited:
How true, I have gotten comments on "why is there a jumper? ". I reply , would you rather me run that input trace through the VAS ? A jumper is just the poor mans double sided layout.
They make dandy test points , too.I now see the PCB traces , lengths and thicknesses as additional components.
OS
Hi Chris & John,
I'm glad we share the same opinion about that chip (that's why I figured out a better solution, see my website), but it is beside my point.
My question was: the bias current is proportional to the abs. temp. [...] should this be regarded as asset or as flaw?
Cheers,
E.
Hi Arthur,
The best I've done with a BJT amplifier is 0.005% THD-20 at 20 kHz into 8 ohms, but I was not really trying that hard or doing anything fancy. The amp used two pair of ThermalTrak devices with RE=0.22 ohms and a gain crossover frequency of about 500 kHz with straight Miller compensation. The amp had 50V rails.
Cheers,
Bob
Hi Bob,
No too shabby for "not really trying".
Hi Edmond,
I haven't a clue how well the bias is tracking in that chip. Due to it's performance, I was not curious to look at it in any detail. Given it's position in the circuit, it seems to me that it can't be tracking bias properly simply because of the way that part affects the sound quality. If it were executed properly in that chip, the answer to your question would probably be a "yes, an asset". That would be my gut feeling at any rate.
Hi OS,
-Chris
No too shabby for "not really trying".
Hi Edmond,
I haven't a clue how well the bias is tracking in that chip. Due to it's performance, I was not curious to look at it in any detail. Given it's position in the circuit, it seems to me that it can't be tracking bias properly simply because of the way that part affects the sound quality. If it were executed properly in that chip, the answer to your question would probably be a "yes, an asset". That would be my gut feeling at any rate.
Hi OS,
Yes, exactly. That's also why jumpers can be superior to a double sided PCB. Using one side as a divided ground plane, insulated jumpers might be a better way to go. That gives you a ground plan and high capacity current delivery routes. In other words, a three sided PCB with real thick copper in one layer. Adding a heavy ground wire / buss along high current return paths might also improve performance, depending on what you routed on that PCB, and how you routed it.
-Chris
Not too shabby , simple is better (straight single pole miller) I always try for .002/5% in simulation ,at least for a BJT OPS. The last one I had tested by a friend at Oak ridge labs really did .02%/ 20k/ 120w. Real world performance is harder to predict even with these excellent on- semi devices (mje15032/33 -4 X njw0281/0302 -EF2).. still a great deal at $13.60 for a full 8 device OPS.
OS
Hi, chris , right?
What I meant was the parasitic capacitance in a long parallel trace or a small inductance in a obtuse routing just to get around the need for an evil jumper. In some commercial offerings, you see them "countercompensate" for shortcomings in their layout with small pF capacitors.
Grounding for heavy rail returns should be VERY short, fat and separate from the sensitive input stage ground , beyond the normal 10-22R resistor...maybe even to the point of a separate ground return scheme (my G1 / G2 setup). I am shooting for the "master's" .005% on the next one(I am trying!
). Your point of wires on the busses should be observed more here on the forums , I see a many 500 watt "arc welder DIY's" with 1/8" -1 oz Cu busses running 10 pair OPS's... OS
Hi Edmond,
I like the LT1166 and had good results with it when I tried it several years back. However, I know that a lot of people have not been successful in achieving low distortion with the device, and the app notes supplied by LTC do not show how to accomplish that. I've got a section in a chapter in my book showing how to use the LT1166 and get good performance.
The LT1166 operates by monitoring the voltage eacross each RE and enforcing the following law in real time: VRE1 * VRE2 = 0.0004. At idle, this comes to 20 mV across each RE. Enforcement of this law results in a non-switching output stage. The LT1166 achieves this by acting as a dynamic bias spreader, essentially used in place of the traditional Vbe multiplier.
The key to using the LT1166 for low distortion is to NOT use its input. You just float it with the signal. It is powered by the VAS standing current, of which the LT1166 needs a minimum of 4 mA. The LT1166 also needs a minimum of 4V across it; this means it may not work with many BJT output stages without some modification.
One must recognize that, as a consequence of the bias law, the voltage at either end of the LT1166 bias spreader will be nonlinear; this has implications for how feedback compensation is employed; if it is not done right, the LT1166 circuit will cause distortion.
The LT1166 also has stability issues, when it is used in the way to achieve low distortion. Because it senses VRE to establish bias spread, we have what amounts to a common-mode feedback loop, which must be compensated. I had to take some measures to properly stabilize this common mode feedback loop that are not covered by Linear Technology. Suffice it to say that the internal compensation of the LT1166 is not adequate for all applications.
The LT1166 is a good, clever part that solves a nagging problem (including dynamic bias stability and thermal lag distortion in the output stage), but it is more difficult to apply successfully than a simple Vbe multiplier, for sure.
Cheers,
Bob
Vt = KT/q
Hi Bob,
I'm looking forward to reading that chapter (and the rest of your book, of course).
Now we come to my original point: At idle, this 20 mV across each RE applies only at room temperature. As a matter of fact, Vre varies with the absolute temperature (of the LT1166, of course): Vre ~= KT/q * ln2
So, if you don't mind, I repeat my question: should this temperature dependence regarded as asset or as flaw?
Indeed, the input transconductance stage is a disaster, as wrong as can be.
Exactly! However, this not only applies to the LT1166, but also to many other dynamic (fast acting) auto bias circuits.
I guess you've put a series RC across pin 1 & 4, i.e. an additional lead-lag compensation, right?
I fully agree with that and I like to add that this chip isn't very fast either. Regrettably, there is nothing we can do about the latter.
Cheers,
E.
Hi Bob,
I'm looking forward to reading that chapter (and the rest of your book, of course
).Now we come to my original point: At idle, this 20 mV across each RE applies only at room temperature. As a matter of fact, Vre varies with the absolute temperature (of the LT1166, of course): Vre ~= KT/q * ln2
So, if you don't mind, I repeat my question: should this temperature dependence regarded as asset or as flaw?
Indeed, the input transconductance stage is a disaster, as wrong as can be.
Exactly! However, this not only applies to the LT1166, but also to many other dynamic (fast acting) auto bias circuits.
I guess you've put a series RC across pin 1 & 4, i.e. an additional lead-lag compensation, right?
I fully agree with that and I like to add that this chip isn't very fast either. Regrettably, there is nothing we can do about the latter.
Cheers,
E.
According to Dale Eagar's "Design Note" improper use of the LT1166 can result in "consternation and hair loss". http://cds.linear.com/docs/Design Note/dn126.pdf
Bob,
The LT1166 assumes that Re1=Re2 doesn't it. Then it can conclude that when Vre1=Vre2 (dynamically) Iout must be zero.
However, common Re resistors are often 10% tolerance, so I think this could then cause the LT1166 to establish a wrong operating point.
Did you see anything of this in your tests Bob?
jan didden
The LT1166 assumes that Re1=Re2 doesn't it. Then it can conclude that when Vre1=Vre2 (dynamically) Iout must be zero.
However, common Re resistors are often 10% tolerance, so I think this could then cause the LT1166 to establish a wrong operating point.
Did you see anything of this in your tests Bob?
jan didden
Hi Jan,
Hi Bob,
I didn't use the inputs of the LT1166 either. In my mind, it was replacing the Vbe multiplier - so I used it like that. The commercial amp I got in later on was designed and built in Italy (from memory on that). It also used the LT1166 as a Vbe multiplier, ignoring the input pins. I found it was interesting that someone else used it the same way I had. There must be a cosmic intelligence at work here!
You seemed to have covered a lot of things that I've been wondering about in your book there Bob. That should make it a very complete text on current audio design ideas. How on earth did you get all this stuff together in one place? Or, how long have you been working on this book?
-Chris
That's an interesting point. On a prototype, I do tend to match these parts, but that normally only weeds out the one or two that are unlike the others in a batch. Most are very close if you buy them in number at the same time. Depends on where you get them too I guess.
Hi Bob,
I didn't use the inputs of the LT1166 either. In my mind, it was replacing the Vbe multiplier - so I used it like that. The commercial amp I got in later on was designed and built in Italy (from memory on that). It also used the LT1166 as a Vbe multiplier, ignoring the input pins. I found it was interesting that someone else used it the same way I had. There must be a cosmic intelligence at work here!
You seemed to have covered a lot of things that I've been wondering about in your book there Bob. That should make it a very complete text on current audio design ideas. How on earth did you get all this stuff together in one place? Or, how long have you been working on this book?
-Chris
Hi OS,
How's that for hairy? The CA-7 spec is pretty low already, I don't have the numbers with me at the moment. I see on the web that the spec was -110 dB in the mid gain position.
The customer? He turned everything off in the house when listening to music. Fridge, hot water tank, furnace - everything. Running the PA-7 amps in bridged mode. You just can't win when people go to extremes like that. I'll bet he could hear the house shifting all the time too!
-Chris
Yep. Hi again. Hope you're keeping well.
I agree with all your points there. In fact, I believe there is value in having various "grounds" (when is a ground, not a ground?) for various uses. A "dirty" ground for HF bypass use, output zobel return, signal and a few others would help get that last bit of noise down. In a Nakamichi CA-7 preamp I serviced years ago, we had a customer complaint that one channel had more noise than the other. Well, both channels exceeded specs by a few dB, but one was far better. I was forced to correct the situation by the area rep (how's that for a warranty situation - re-engineering the damn thing). I did it though. It required some shielding, and I even ran a ground inside a shielded cable, outer shield grounded at the supply only. That solved it and both channels were performing about 10 dB better than spec.
How's that for hairy? The CA-7 spec is pretty low already, I don't have the numbers with me at the moment. I see on the web that the spec was -110 dB in the mid gain position.
The customer? He turned everything off in the house when listening to music. Fridge, hot water tank, furnace - everything. Running the PA-7 amps in bridged mode. You just can't win when people go to extremes like that. I'll bet he could hear the house shifting all the time too!
-Chris
Here is my data, though using LME49811 driver chip. But I payed much attention to layout and wiring to get the performance. My rails is +/- 40. MJE15030/31 driver and two pairs of Sanken 2sc3624/sa1925 EF, RE=0.11 ohms form the output stage. Single-pole compensation is used.
Attachments
Hi Bob,
I'm looking forward to reading that chapter (and the rest of your book, of course
Now we come to my original point: At idle, this 20 mV across each RE applies only at room temperature. As a matter of fact, Vre varies with the absolute temperature (of the LT1166, of course): Vre ~= KT/q * ln2
So, if you don't mind, I repeat my question: should this temperature dependence regarded as asset or as flaw?
Indeed, the input transconductance stage is a disaster, as wrong as can be.
Exactly! However, this not only applies to the LT1166, but also to many other dynamic (fast acting) auto bias circuits.
I guess you've put a series RC across pin 1 & 4, i.e. an additional lead-lag compensation, right?
I fully agree with that and I like to add that this chip isn't very fast either. Regrettably, there is nothing we can do about the latter.
Cheers,
E.
Hi Edmond,
If anything, I think the positive TC of the 20 mV number is a good thing; it certainly would be in the right direction if it was reflective of the output transistor temperature, since Vq = Vt has a similar positive temperature coefficient. However, over the temperature ranges that the LT1166 can be expected to operate (essentially board ambient), the temperature change as a percentage of absolute temperature is quite small, so I don't think it is much of a factor one way or the other.
It is notable, however, that if you take the spec sheet at its word, there is quite a large tolerance on the 20 mV, about a 2:1 p-p range. This is not desirable, if really true. The ones I looked at were very close to 20 mV. If this were a problem, I would not rule out adding a trim pot. I know, it sounds crazy, but eliminating the trim pot is not the only potential advantage to a well-executed 1166 circuit - non-switching and dynamic temperature stability.
Yes, I essentially introduced a series R-C across the device. BTW, as a shunt bias spreader, the device is more than adequately fast, and in fact this is part of the reason the compensation of the common mode loop is necessary.
Cheers,
Bob