Discrete mic pre project: proto PCBs have arrived.

[jdbakker]

Pay attention to your resistors. I've found very high 3rd harmonic in many parts. The Susumu 0805 RR12P are excellent.

Agreed; I've used them in the past in low-noise analog designs and am planning to spec them for this project as well. Pity that 1206 thin film parts are hard to get, but I expect that most people will be able to handle 0805s just fine as long as I don't pack them too densely.

Delivery on many thin film in values we typically need e.g. 6K81 1/2W are problematic. The thermal mass of many 0603 and 0805 parts are small enough to create thermal effects. Be very, very careful. SMT may be "cool," but not necessarily better.

Yes. On one hand I'm looking at reducing the number of line items, like possibly replacing a 300-Ohm resistor with three 100-Ohm parts if the former value is used only once in the design (as I've found that both for manual assembly and P&P it's faster and cheaper to place more parts if they have less different values). On the other hand there are a few spots where I am brute forcing low voltage noise by choosing low-resistance feedback networks, and for those I may just go for through-hole after all, both for the reasons you mention and for the simple fact that quality low noise SMD power resistors are just silly expensive (although the jury's still out on MELFs).

JD 'decisions, decisions' B.

 

[Svart]

MILFs?  I like MILFs..  ;D

Melfs on the other hand are ok too.

 

[jdbakker]

Melfs on the other hand are ok too.

You say that because you work in Design, not Assembly. Assembly folks tend to hate MELFs as they usually don't have a flat top (no good for most P&P machines) and even with the recommended slotted footprint they are really good at rolling away before they're soldered down. This is also the main reason why I'm not sure we should use 'em in this design (availability being another).

The thermal mass of many 0603 and 0805 parts are small enough to create thermal effects.

One thing that helps is to have mucking big copper planes connected to them. In a previous project we found that an 0603 bridging a narrow gap between two outer layer planes could sustain ~1W for days on end without burning or even drifting outside of its specced range. Inspection with an IR camera showed that the part remained reasonably cool (30ish C above ambient, if my memory serves me well).

JDB.
[I guess the alumina board substrate didn't hurt either]

 

[Moby]

MILFs?  I like MILFs..  ;D

Melfs on the other hand are ok too.

Yes, I vote for SMD and MILF's  ;D

 

[Svart]

Melfs and manufacturing: yes some hate them.  I've seen some companies use a stickier version of solder paste to hold them in place before baking the boards in IR ovens but this usually costs more and isn't worth the cost if you can find a different package.

 

[brazilianwonderboy]

Absolutely. That's one of the reasons that I still intend to use traditional output transistors like the BD1xx/MJE1xx-series.

I was thinking about just this subject recently and started looking for SMD alternatives to the MJE-171. I came across the BCX53 as a possible alternative. What do people think? I'm not an expert of transistors so I'm sure I've missed something, but from the datasheets they seem relatively close (or at least the closest match I've found so far).

 

[jdbakker]

I was thinking about just this subject recently and started looking for SMD alternatives to the MJE-171. I came across the BCX53 as a possible alternative. What do people think?

Should work. The datasheet suggests that it uses the same die as the BC640, which has been said to be the same die as the now out-of-production Philips BD140, and I've always been happy with those.

JDB.
[never used the BCX53 myself, and none of the listed packages are particularly friendly for the novice SMD DIYer]

 

[brazilianwonderboy]

The Susumu 0805 RR12P are excellent... [..] ... The thermal mass of many 0603 and 0805 parts are small enough to create thermal effects.

So are you saying that 0603 and 0803 parts are in general bad, but with the exception of the Susumu's? Or that 0603/0803's seem to be the best option in general? Sorry for my confusion.

Should work. The datasheet suggests that it uses the same die as the BC640, which has been said to be the same die as the now out-of-production Philips BD140, and I've always been happy with those.

Cheers for the response, I guess the next test will be just to try it out and see how well it operates. I'm gunna get some I think and have a tinker.

 

[Svart]

If you want the best, then go with the Susumu, which are thin film.  Otherwise you probably won't hear too much difference in general usage with thick film or metal films.

 

[jdbakker]

The Susumu 0805 RR12P are excellent... [..] ... The thermal mass of many 0603 and 0805 parts are small enough to create thermal effects.

So are you saying that 0603 and 0803 parts are in general bad, but with the exception of the Susumu's? Or that 0603/0803's seem to be the best option in general?

The resistance of all real-world materials has a temperature coefficient that is not exactly zero. Clever materials engineering may create stuff with a resistance tempco of close to zero, but in real life any resistor you can buy/build will change its resistance when its get hotter/colder.

When you run a current through a resistor, it will heat up (P=I2R). So, any real-life resistor will through self-heating have a current-dependent resistance.

When you run a current through a resistor with a small thermal mass (which will be the case for 0603/0805s), self-heating will cause a rapid change in temperature and thus resistance. If this rate of change is comparable to the signal frequency range you're interested in, you'll get distortion. Imagine what happens in an amplifier where the feedback network has a different resistance at signal peaks than at signal nulls, for example.

Fixes:
- use resistors with a large thermal mass (this usually means 'physically big')
- use resistors well below their rating. Less dissipation = less resistance change
- use resistors with a low temperature coefficient. Sadly not all manufacturers spec this

These fixes can be used individually or combined, and while none will eliminate the problem, in an active circuit you can make them small enough that they don't dominate. Also keep in mind that only the AC component of the current through a given resistor matters; DC will shift the resistor somewhat off its room temperature value, but that's generally no big deal in audio circuits.

Personally I often run into power dissipation limits with SOT-23 parts though.

Yeah, I have the same problem. +/-24V rails * 5mA = awfully close to the dissipation limits of SOT-23 parts. I'll see if I can use a few TO-92 parts in strategical places to deal with that issue (cascodes would be a natural choice, and I have a few of those already).

JDB.

 

[Rochey]

Quote:

When you run a current through a resistor, it will heat up (P=I2R). So, any real-life resistor will through self-heating have a current-dependent resistance.

//

Does that mean that if we run into high impedance, the current will be low, and therefor the resistors won't heat up much.

Just thinking about places in the circuit where I *must* specify those special resistors, and where I can just specift 1 percent thick metal.

Btw - some customer feedback was that they used 1 percent thick film smd's to get over 120dB from their ADC. I guess it's all a balance of vlues, performance and cost...

 

[jdbakker]

Just thinking about places in the circuit where I *must* specify those special resistors, and where I can just specift 1 percent thick metal.

Meh. If you're not competing with the Mainland, might as well standardize on them for SMD audio work. In reel quantities they're about a cent each. Stocking two different parts, having to think about which one to use and P&Ping an extra part will cost you much more than you could save.

Btw - some customer feedback was that they used 1 percent thick film smd's to get over 120dB from their ADC. I guess it's all a balance of vlues, performance and cost...

And luck. Noise-wise I've had both good and bad experiences with regular thick film SMD parts. The thing is that very few manufacturers specify excess noise for thick film parts, and you know that means you'll be holding the bag if the new batch suddenly is a lot noisier than the old one. I believe at least Susumu does have such a spec for their thin film parts, not sure about Vishay and the others.

JDB.

 

[Rochey]

. In reel quantities they're about a cent each. Stocking two different parts, having to think about which one to use and P&Ping an extra part will cost you much more than you could save.

Let's put that in perspective... I pay about $4 for 5000 1% 0805 resistors. (thick film)
5000 resistors at $0.01 each is $50

Add to that, that you need multiple values, and a small production run could get costly very quickly without selectively choosing where you use the expensive ones, and where you use the cheapies.

Newark has them for $0.06 each... in low quantitiy. That may be worthwhile... Even digikey does them at $0.09 each in cut tape.
Thats worth it for limited use, but if you specify them all over your board (even for things like digital logic) then the costs for that become ridiculous too.

 

[jdbakker]

In reel quantities they're about a cent each. Stocking two different parts, having to think about which one to use and P&Ping an extra part will cost you much more than you could save.

Let's put that in perspective... I pay about $4 for 5000 1% 0805 resistors. (thick film)
5000 resistors at $0.01 each is $50

I should have qualified that statement. The $50/5k price is at Digi-Key; I looked at prices for 1k 1% (or better) parts. Cheapest thick film 1% reel at Digi-Key is $13/5k. But that's Stackpole, and for some reason my assembler doesn't like those. Next up is Rohm (generally OK) for $21/5k.

My assembler prefers to source his own parts as he gets better prices that way, I'd expect the Susumus to come down too. Where do you get $4 reels, BTW?

Add to that, that you need multiple values, and a small production run could get costly very quickly without selectively choosing where you use the expensive ones, and where you use the cheapies.

Still disagree from a stock management POV. In any design you want to (a) minimize the number of different parts and (b) stick to common values. I dunno what your experience is, but when hand assembling stuff I spend much more time switching between parts than actually placing parts. Come to think of it, on a mixed board I place about 150-200 parts per hour, and my time isn't free. Even at McDonald's wages resistor cost gets lost in the noise.

Newark has them for $0.06 each... in low quantitiy. That may be worthwhile... Even digikey does them at $0.09 each in cut tape.
Thats worth it for limited use, but if you specify them all over your board (even for things like digital logic) then the costs for that become ridiculous too.

For digital logic you generally don't care about the exact value, might as well reuse whatever you happen to have elsewhere. On my watch board I used 470k pull-ups and 1.5k LED current limiting resistors as I needed those values in the analog circuitry.

JDB.
[on the other hand, I have a few reels of 100n NP0 ceramic capacitors, and those will never get re-used for generic digital decoupling as they are rather more expensive than the thin film resistors. Plus they're 1206, a bit bulky for modern digital stuff]

 

[jdbakker]

OK, I've finished the design for a first version of the input amplifier. Two of these together in the familiar Green/Cohen-configuration (plus possibly a third to cancel common-mode signals) form the first stage. Gain is fixed at 26dB, noise figure is about 1.5dB and it accepts input signals of 3.5Vpeak-peak before clipping so it should be able to handle 0dbV signals without a pad.

Yes, the amp has a lot of parts, but keep in mind that most of those transistors are <5 cents each at the usual distributors so this amp isn't going to cost more than a regular DOA (less even if you don't use a matched pair as input transistors).

The output of this front-end is fed through a variable attenuator to a +20dB differential in/out amplifier. If I get some time I'll draw up a schematic.

JDB.

 

[jdbakker]

Update: finished the first revision of the differential I/O amps that will be used as the second and third stage of the mic pre. Comments are very welcome.

Next up: gain structure. I could definitely use some input there, will try to find some time to draw up that block diagram I promised earlier.

JDB.

 

[Rochey]

Opa1612?

 

[jdbakker]

Opa1612?

...huh?

Do you mean the OPA1612, a dual audio op-amp, or the OPA1632, a diff in/out amp?

Both look like nice parts (I've used the OPA1632). Neither will swing +/-20V into 100Ω (single-ended) or +/-40V into 200Ω (differentially), like the linked designs do.

Neither is terribly ... discrete, which sort of is the subject of this thread, right?

JD 'puzzled' B.

 

[jdbakker]

Gain structure stuff is posted here. Let me know what you think.

Thanks,

JDB.

 

[Rochey]

Opa1612?

...huh?

Do you mean the OPA1612, a dual audio op-amp, or the OPA1632, a diff in/out amp?

Both look like nice parts (I've used the OPA1632). Neither will swing +/-20V into 100Ω (single-ended) or +/-40V into 200Ω (differentially), like the linked designs do.

Neither is terribly ... discrete, which sort of is the subject of this thread, right?

JD 'puzzled' B.

missed the "discrete" bit

Yes, the 1612 was what I had in mind. We have many boutique mic pre guys evaluating that part for mic pre's. 1nVrtHz is very tempting for many.

/R