Cheap matched transistor pairs

[saint gillis]

  Dear all,
I found these kind of very cheap NPN transistor matched pairs :
http://www.mouser.com/ds/2/115/ds30311-19227.pdf
http://www.mouser.com/ds/2/302/BCM847BV_BS_DS-352886.pdf

  The matching accuracy is not so far from a THAT300 array or an SSM2210 / SSM2212 for instance...

  I'm looking for a matched pair to make an exponential current converter for a VCO, something like that :
http://schmitzbits.de/expo_tutorial/index.html

  So I'm not looking for something for an audio differential receiver... only a stable VC current source

  What do you think of the cheap components above?

 

[JohnRoberts]

Try them and see how they work... in a single package thermal tracking should be decent.

Sounds like they may use two discrete devices mounted into a single package. 10%  beta and 2 mV offset is not bad but not as good as the better single chip monolithic duals (like LM394). Perhaps adjacent parts from a larger die.

I used to like the old (and cheap) CA3086 that was a 5 (npn) transistor monolithic array. The 5 devices on a single piece of silicon could have some weird interactions via the substrate body diodes, but I used tons of them to make log convertors for meters,  compressor side chains, and the like.

My old favorites are obsolete these days.. discrete design is not as popular as it once was.  :'(

JR

 

[Jarno]

For handsoldering rather than reflow soldering, look at bcm847ds that's a bigger package.
They also have a NPN/PNP pair (not matched).
Interesting parts, i've used them in a few designs.

 

[PRR]

> matching accuracy ....  exponential current converter for a VCO

The initial offset is not critical in this application; the trimming process will null that.

We used to use loose parts. In days when "peas from a pod" could be very different.

The essential item in a music Exp VCO is thermal tracking. Moog and ARP tried several schemes, from wrapping in foil to potting.

Two same-lot modern-process devices in one SOT ought to be better than anything we had in the old days.

I advise you, from frustration, to add more thermal insulation to reduce short-term drift. National had a volt-ref they sold in a little plastic snap-case. A spritz of window insulating foam both sides of the PCB (after initial testing!) might be good. Don't work in a school with crappy thermostats.

 

[CJ]

DBX 2001 VCA also used the conductive tape, metalized polyester according to SSL Tech,

 

[saint gillis]

Don't work in a school with crappy thermostats.

 

[Samuel Groner]

Even off-the-shelf single transistors have really tight Vbe matches; I've recently measured several hundred transistors to study this. Most randomly picked transistor pairs have at most some hundred uV offset; it takes time to find 1 mV mismatch, and for certain transistor types typical figures go down even to tens of uV.

While there are no guarantees I'd say any reasonable dual transistor should do as well or likely better. The reason for the rather high maximum data sheet values is probably that they don't test this parameter (which perhaps also accounts for the low cost).

So I suggest you keep an eye on other effects which are very important as well:

* the magnitude of parasitic (base and emitter) resistance;
* the effect of finite beta (and its dependence on collector current);
* the importance of keeping both junctions at the same temperature (minimizing the power dissipation difference of the two transistors, good PCB layout practice);
* the influence of Early effect.

Samuel

 

[JohnRoberts]

Even off-the-shelf single transistors have really tight Vbe matches; I've recently measured several hundred transistors to study this. Most randomly picked transistor pairs have at most some hundred uV offset; it takes time to find 1 mV mismatch, and for certain transistor types typical figures go down even to tens of uV.

While I don't dispute your experience that wasn't always the case. Apparently the production process that made those transistors is very tightly controlled and free of impurities.

It might be of interest to measure a few parts from different production batches and maybe some older parts to see how close those are. (I assume you tested Vbe at a typical operating current.)

While there are no guarantees I'd say any reasonable dual transistor should do as well or likely better. The reason for the rather high maximum data sheet values is probably that they don't test this parameter (which perhaps also accounts for the low cost).

Indeed if singles are that good a dual will also enjoy improved thermal tracking.

So I suggest you keep an eye on other effects which are very important as well:

* the magnitude of parasitic (base and emitter) resistance;
* the effect of finite beta (and its dependence on collector current);
* the importance of keeping both junctions at the same temperature (minimizing the power dissipation difference of the two transistors, good PCB layout practice);
* the influence of Early effect.

Samuel

As usual thanks for sharing, I would never expect that result. Did you also test a few different device part numbers?

JR

 

[Samuel Groner]

Yes, process control probably got better over the past decades.

As usual thanks for sharing, I would never expect that result. Did you also test a few different device part numbers?

So far I've tested four different devices, from three different manufacturers.

Samuel

 

[JohnRoberts]

Yes, process control probably got better over the past decades.

As usual thanks for sharing, I would never expect that result. Did you also test a few different device part numbers?

So far I've tested four different devices, from three different manufacturers.

Samuel

Funny... back in the '80s I got a patent for a cheap (class A) VCA where I used a separate DC feedback loop to accommodate using unmatched transistors in production without factory trims (4,818,951). 

Later we abandoned the approach after dbx started selling their (much better than mine) VCAs to competitors. Class A VCAs had poor SN. 

Sounds like with modern transistors I could just drop them into the circuit unmatched,  and avoid all the bother.

JR

PS: Large geometry low noise transistors also suffered from Vbe offsets, back in the day requiring trims, DC servos, or cap coupled gain pots. 

 

[PRR]

> off-the-shelf single transistors have really tight Vbe matches; ...; it takes time to find 1 mV mismatch

1970s, 10mV was more typical. 1mV pair was unlikely.

> I assume you tested Vbe at a typical operating current.

If this is a music-synth log converter, match must be maintained over 2 maybe 3 decades. As you may be tuning a 50Hz bass-line with additive-synthesis 5KHz overtones of the melody line, few-Cent tracking may be wanted over 1:100 range of current.

As said, parasitics like base resistance skew you at high current, and leakage bends your low notes. hFE curve also matters, though most designs make this second-order. Early is usually ~~nil because Vce is kept tiny and stable. (I do recall nulling 5mV opamp offsets because that much shift of Vce mattered.)

> I got a patent for a cheap (class A) VCA where I ....

lapsed 1993  "failure to pay maintenance fees"   

 

[JohnRoberts]

> off-the-shelf single transistors have really tight Vbe matches; ...; it takes time to find 1 mV mismatch

1970s, 10mV was more typical. 1mV pair was unlikely.

> I assume you tested Vbe at a typical operating current.

If this is a music-synth log converter, match must be maintained over 2 maybe 3 decades. As you may be tuning a 50Hz bass-line with additive-synthesis 5KHz overtones of the melody line, few-Cent tracking may be wanted over 1:100 range of current.

As said, parasitics like base resistance skew you at high current, and leakage bends your low notes. hFE curve also matters, though most designs make this second-order. Early is usually ~~nil because Vce is kept tiny and stable. (I do recall nulling 5mV opamp offsets because that much shift of Vce mattered.)

In the Vbe log convertor for the dB meter in my old TS-1 the Rbb of the small transistors (CA3086) caused a measurable dB error at the top of the range +20dBu.  The Rbb resistance in series with the base emitter diode,  added a small but significant voltage error at the top of the range, while down at -80 dBu not so much.  ;D ;D ;D

> I got a patent for a cheap (class A) VCA where I ....

lapsed 1993  "failure to pay maintenance fees" 

I sure hope they stopped paying maintenance on it, I abandoned it as soon as I could buy better VCAs off the shelf at reasonable prices (THAT corp).

I used to argue with Hartley and advised against him filing for international patent for most (all?) of my inventions. But he liked all the expensive wall paper. A patent just gives you the right to sue, and that makes little sense in relatively small international markets. The only one of my inventions that was really commercial (the FLS LEDs over the GEQ sliders) got ripped off by Behringer and they won in court when Peavey sued them. Technically they "didn't lose" in court, but that's a win when do what they did . 

JR