Ideas

I was thinking about another instrument for audio - an accurate differential dB meter. This might be helpful in making gain or loss measurements. Also, measuring amplifiers with balanced outputs or other non-ground referenced signals.

Old-school VU meters use a very simple circuit with full-wave rectifiers to drive the meter mechanism. They depend on the non-linear behavior of the diodes to create a logarithmic response, but this is not very accurate. Of course, a small VU moving needle meter is not expected to be very accurate but they were fine for things like analog tape decks.

Professional-grade instruments are very pricey since they are so specialized. 

Several functions are required: an input attenuator (perhaps with auto-ranging), a differential amplifier (with adjustable gain), an RMS converter for AC signals, a log converter to get dB readouts, and a display of some sort. The display could be as simple as 7 segment digits or something fancier like an LCD readout (and maybe a touchscreen for controls).

Another useful feature might be to include peak detection.

So I was wondering about a cost-effective way to build one. The heart of the device is the RMS conversion device. Years ago there was a chip from Linear which used thermal techniques to do true RMS conversion. Unfortunately, that chip became obsolete. Some are available on the used market for big bucks.

More recently, I came across the Analog Devices (originally Linear) LT1968. Datasheet is here. This chip has been around for many years. Jim Williams wrote an app note describing various circuits using the devices in the family. It can be used in either single-ended or differential mode, so may provide the differential function out of the box. This may not be the ideal chip for the job, but since it has good bandwidth and low noise may be a good starting point. What is good about it is that it uses a delta-sigma conversion technique so it is accurate for different waveforms. Another possibility is the AD736/AD737, but it uses the "average responding" technique, meaning full-wave rectification, which is only very accurate for sine wave inputs. That may be fine, but it is not great for things like noise measurement. 

There are several log converter ICs on the market, but many appear to be aimed at the RF market. One possible candidate is the TI (previously Burr-Brown) LOG104 chip. Datasheet is here. It operates on current inputs, so would require V/I conversion, but it can compute log ratios as well, making it more versatile. Another possible chip is the LOG114 which appears to be fully differential and can be used for voltage inputs. Most of the applications I have seen for these devices involve photometrics. These devices can be problematic when implementing practical circuits since they need to operate over a wide range, and become inaccurate at the extremes, so that needs to dealt with.

So there are some possibilities for prototyping the core of such a device.

It might be feasible to make the instrument "intelligent" by adding e.g., a Raspberry Pi or Arduino and display. It could then have a display configured in software and provide logging capability for measurements - that would require some extensive coding and may be overkill, but cool.

The attenuator I used in my preamp project had one flaw: the relays are driven by a linear binary code sequence and therefore,  at any given volume step, anywhere from one to 8 relays are changing state. This results from a great deal of clattering on some of the steps. The worst case is that ALL the relays will change on particular steps.

There is an old encoding technique called Gray Code encoding where only one-bit changes with each step. I came across this post in DIY Audio by contributor jpk73 who had the same idea and simulated it here. The binary to Gray Code conversion is fairly simple - it is done with a set of exclusive or gates controlling a set of DPDT relays. These should be latching, as is the case with my existing setup.

There is an older patent, now expired, covering this idea as well.

It would be much nicer to have only one relay changing state as the volume is ramping up or down.

This should be trivial to implement since the conversion is done by just a few gates.