Just feed three different chains of LM3915's. Feed one chain of two directly, add a one dB pad before a second chain of two in series and a two dB pad before a single LM3915. If those IC's are accurate you end up with 50 LED's with one dB per step after you mount the LED's appropriately.
Interesting idea, enough so that I read the datasheet. :-) The LM3915 is not accurate enough for 1dB step for the whole range, but it is for the top 10dB. By my experience, unless you are building a measurement tool, a VU meter needs to:
- show the difference between a really faint signal and no signal: Only a few steps under -48dB, and none under -60 or -70dB.
- 3dB/step to -20dB
- 1dB/step over -20dB. Half-dB steps are nice, but luxury.
So, the above idea is sound, with some limits.
--
[ Some random thoughts about VU meters:
Nowadays with digital equipment, a peak hold function (that LM3915 does not have) is useful on inputs: You want to know if you overshoot a recording and by how much. At the very least, a holding "clip" indicator I a must, if recording. Even in live sound, if you heard the clipping you'd like to know what channel caused it.
The more channels you have, the more individual channel cost matters. Turned out that for me, the most economical solution was is a distribution mixer, a multichannel soundcard, a PC, an big monitor and
http://www.darkwooddesigns.co.uk/pc2/Multi.html software.
Numerical values (i.e. a digital meter) are not needed, unless you want to do measurements.]