What exactly are complementary matched pair transistors?

[rfengg]

Hi All,

Came across the NJW0281G (NPN) and NJW0302G (PNP) transistors from ON Semiconductor in a repair job in which the common data sheet for the complementary pair claims to "NPN/PNP Gain Matching within 10% from 50 mA to 3 A"
However what is confusing for me is that , the data sheet also says that hfe can vary everywhere 75 and 150 for any of the 2 devices.....if so, how is this a matched pair? What parameters exactly do you look for when selecting a matched complementary pair if one exists due to the variations in PNP and NPN devices? Do we really have to sit and measure hfe's for many transistor pairs if they have to be used in say, the finals of a audio amp and is there a significant advantage to this especially in terms of distortion for dc coupled amps?
Page 4 of the data sheet gives a plot of "NJW0281G Current Gain BandwidthProduct" and "NJW0302G Current Gain BandwidthProduct" which look quite similar, but given the variations of hfe between 75 and 150, how is this guaranteed? 

[M4trix]

Real matching is done by measuring the Vbe. If the hfe is in the close range, the better. There is an article somewhere on the internetz where Crown audio has explained the difference between matching by Vbe or hfe. If you're building an amp, I wouldn't bother much about matching the PNP with the NPN ones. You won't hear the difference anyway. More important is to match NPN groups and PNP groups.

NPN/PNP Gain Matching within 10% from 50 mA to 3 A

Well, one would say, matching is not even necessary then. Not true. That's why I always match my transistors.

[Kleinstein]

It depends on the amplifier circuit if NPN to PNP gain matching makes a significant difference. In most circuits it does not make a difference. Usually good designs don't rely on gain matching in the output stage. It is more with the input stage differential pair that VBE and gain (for a pair of same type) matching may be relevant to get a small offset.
For the NPN to PNP matching is usually for the gain, if done at all. It is rare to need VBE matching for a NPN/PNP combination.

For pair from the same type matching is more often wanted for VBE. For parts from the same batch this usually also leads to reasonable matching of other parameters like the gain and Early voltage.

Matching is usually done by measuring the relevant parameter for a set (e.g. 10)  of parts and than look in the data for the pairs that are close. It does not need than many parts, as the number of possible pairs goes up like N*(N-1)/2 with the number of parts. As an example it only needs some 25 persons to have a good chance that 2 have the same birthday.