Meter with two consumer-side connections

[SimonH2]

there is a lot of expertise out there, some ready and waiting to be 'bought in', even if they don't have the requires skills in-house

Indeed - but that costs, and there's a trade-off between paying a little extra for the dedicated chip per meter built, vs paying a big chunk up front for bought in expertise and amortising it across however meters they hope to make before the design specs get changed. There's also the tradeoff between having the processor doing "soft" real-time tasks, and the harder task of having it do "hard" real time tasks which it must do if you do the sampling and maths in software. For a company with roots in "mechanical" stuff (such as Farraris disk meters) then they may be "cautious" with software and use a dedicated measurement chip, but if the company has a lot of expertise in software (and in particular, hard real-time systems) then they might be more prone to doing it in software.
Don't forget that as well as working as a meter, the new "smart" meters also have to handle encyption and comms - thus loading up the number of variable CPU load tasks it has to do.
There is no easy answer here - and without knowing all the facts, you can't really say whether any manufacturer's approach is "the best".

 

[John D v2.0]

This reminds me of the results screen at the Olympic pool that usually shows a time clock during public sessions. Very useful for timing yourself, and clearly the front end samples an accurate clock chip once a second and renders the clock face.
Works well mostly, but the front end timing is not accurate, so every so often you are waiting to start on a specific time and it just skips that second.
And all this from something specifically there for timing yourself at the best competition pool in the country!

 

[JohnW2]

There are advantages both ways - but I suspect that the dedicated energy monitor chip is more accurate and needs less (and less complex) software.

There are always pros and cons If the actual 'energy measurement' is done ('in hardware') by an off-the-shelf chip, then, yes, as you go on to say, the (separate) processor obviously only has to have relatively simple software to do the more 'trivial' things (display, storage, comms. etc.) - fairly standard programming which does not require any specific expertise in DSP (or even an understanding of the electrical principles involved).

Of course, the interesting thing is that, despite the above argument in favour of use of the dedicated chip, the one example we have heard of in this thread is of a meter which apparently does not use such a chip. This may indicate that (component) cost considerations have over-ridden the argument - although, as you say in your subsequent post (to which I'll reply) the cost situation is more complex than just component cost.

Doing it in software needs "quite a bit of care", in particular timing is critical as you need to keep the read-calculate-accululate loop running at a constant rate regardless of what else the processor is doing.

It certainly needs "quite a bit of care", but I'm not convinced that the specific point you mention, in itself, would be a major issue, particularly given the high sampling rate.

Kind Regards, John

 

[JohnW2]

Indeed - but that costs, and there's a trade-off between paying a little extra for the dedicated chip per meter built, vs paying a big chunk up front for bought in expertise and amortising it across however meters they hope to make before the design specs get changed.

Quite - but, as I've just written, the one meter we've heard about in this thread appears to have decided not to use the dedicated chip.

There's also the tradeoff between having the processor doing "soft" real-time tasks, and the harder task of having it do "hard" real time tasks which it must do if you do the sampling and maths in software. For a company with roots in "mechanical" stuff (such as Farraris disk meters) then they may be "cautious" with software and use a dedicated measurement chip, but if the company has a lot of expertise in software (and in particular, hard real-time systems) then they might be more prone to doing it in software.

Again, agreed. As I recently wrote, if the energy metering is done by an off-the-shelf chip, the only programming necessary is fairly standard stuff, not requiring 'unusual' specific skills or knowledge of the engineering principles involved.

There is no easy answer here - and without knowing all the facts, you can't really say whether any manufacturer's approach is "the best".

Indeed so. ... and we don't know (except in one case) what decisions the manufacturers have reached.

Kind Regards, John

 

[ban-all-sheds]

... accululate ...

 

[JohnW2]

... I was thinking, probably almost subconsciously, that implementing (presumably fairly high precision) multiplication of two binary numbers by 'hard-wired logic' would be 'too much like hard work' (I cannot forget building a very rudimentary 'calculator' out of discrete logic elements in my mis-spent youth!), and that they would therefore probably have included some very simple standard programmable processor 'module' on the chip, along with a set of hard-wired (or burnt-in, or whatever) instructions to tell the processor how to do the multiplication ....

On reflection, I don't think it would be the 'multiplication' which would be the biggest headache - implementing the digital low-pass and high-pass filters 'in hardware' would probably be much more challenging. For someone with an adequate understanding of the maths and engineering, it's relatively straightforward to implement such filters in software, but I personally wouldn't have a clue where to start trying to do it in hard-wired logic - and I imagine that, even for someone who did 'have a clue', it would end up pretty complex.

However, some clever person(s) have clearly done it!

Kind Regards, John

 

[bernardgreen]

Filtering in hardware introduces phase shifts in the signals so these unwanted phase shifts have to be corrected if subsequent calculations involve V x I across the full power factor range.

Digital sample and hold on V and I then digital processing is far "easier" than a hardware ( capacitor, inductor, resistor ) filter.

 

[JohnW2]

Filtering in hardware introduces phase shifts in the signals so these unwanted phase shifts have to be corrected if subsequent calculations involve V x I across the full power factor range.

Indeed, and that's probably why the AD71056 has a 'phase corrector' after the current ADC. However, I think you may be misunderstanding, if by "fitering in hardware" you mean analogue filtering - something which no-one has suggested. The discussion has been about the difference between filtering of the digital data by (a) hard-wired logic ('hardware') or (b) by a software-instructed processor.

Digital sample and hold on V and I then digital processing is far "easier" than a hardware ( capacitor, inductor, resistor ) filter.

As above, no-one has ever suggested an analogue filter. In any event, the requirement is not to filter the waveforms being sampled but, rather, the digital conversions thereof, to remove the 'DC offset' in the output from the current ADC.

Kind Regards, John

 

[bernardgreen]

It was many many years ago and I had no direct involvement but analogue filtering was considered necessary in "electronic" meters. Something connected with pulsed remote control signals for lamposts along the mains, or remote control of Off Peak metering or was it ( ultra ) low frequency radio transmissions for comminications with submerged submarines ( such as a UK version of Project Sanquine ) affecting the waveform of electrical supplies.

 

[JohnW2]

It was many many years ago and I had no direct involvement but analogue filtering was considered necessary in "electronic" meters. Something connected with pulsed remote control signals for lamposts along the mains, or remote control of Off Peak metering or was it ( ultra ) low frequency radio transmissions for comminications with submerged submarines ( such as a UK version of Project Sanquine ) affecting the waveform of electrical supplies.

There's obviously all sorts of 'noise' in electrical supplies (not the least being 'data' which is deliberately there, as well as induced RF), particularly these days, and one obviously has to make sure that does not adversely affect the metering electronics. However, I think that is a totally different matter from the issues being discussed her - as I said, the (digital) filter in the the output of the current ADC in the AD71056 is a high-pass one (hence would do nothing for the sort of 'noise' you describe), and is there to remove the DC offset component of the output of the ADC (not anything to do with filtering the mains waveform).

Kind Regards, John

 

[endecotp]

On reflection, I don't think it would be the 'multiplication' which would be the biggest headache - implementing the digital low-pass and high-pass filters 'in hardware' would probably be much more challenging.

There are some clever tricks that you can do to greatly reduce the amount of logic needed.

A naive implementation might have an ADC with a 10-bit-wide output; then your digital filter needs a lot of 10-bit adders and multipliers. The gate count quickly adds up.

An alternative would be to use an ADC with a narrower output (e.g. 4 bits) and a higher sample rate; then you need only 4-bit adders and multipliers. At the output you then downsample to a lower sample rate with more bits, averaging the values. If you do this right you get the same numbers as the naive implementation at much less cost.

Taking this to its conclusion, you can use an ADC with a 1-bit output. This is a sigma-delta converter. In effect, the probability of the output being a 1 indicates the input voltage. Your digital filter is then made up of "1 bit" arithmetic; specifically, you need to multiply the fixed filter coefficients by 0 or 1, which is trivial.

 

[JohnW2]

A naive implementation might have an ADC with a 10-bit-wide output; then your digital filter needs a lot of 10-bit adders and multipliers. The gate count quickly adds up.

Quite so.

An alternative would be to use an ADC with a narrower output (e.g. 4 bits) and a higher sample rate; then you need only 4-bit adders and multipliers. At the output you then downsample to a lower sample rate with more bits, averaging the values. If you do this right you get the same numbers as the naive implementation at much less cost. ... Taking this to its conclusion, you can use an ADC with a 1-bit output. This is a sigma-delta converter. In effect, the probability of the output being a 1 indicates the input voltage. Your digital filter is then made up of "1 bit" arithmetic; specifically, you need to multiply the fixed filter coefficients by 0 or 1, which is trivial.

That sounds potentially very clever, but I'm struggling a bit to get my head around it, so may have to try some simulations in an attempt to convince myself!

In particular, given the very wide range of possible currents being metered, I can't see how it could work with a 1-bit (or even 4-bit) current ADC - since, if it could 'cope' with the highest possible currents, during 'normal use' the output of the current ADC would nearly always be (binary) zero, throughout all of each and every cycle, so that the instantaneous power calculated (by multiplication) at every sampling point would also be zero, and hence the integrated power over any number of cycles would also be zero. What am I missing?

Kind Regards, John

 

[ban-all-sheds]

Whether they are still saveur-du-mois IHNI, but going in the other direction, bitstream, which is 1-bit DACs with oversampling, is/used to be used in domestic CD players and standalone DACs.

 

[JohnW2]

My simple mind is still somewhat struggling with the concept of 1-bit DACs in most situations - since they surely must change any input waveform into a square wave OR (if the amplitude of the input signal never crosses the 0/1 changeover threshold) DC (at '0' or '1' level)? It was the latter of those situations which I was thinking would normally be the case with the current DAC in a kWh meter - with the output of the DAC always being '0'. As I asked, what am I missing?

Kind Regards, John

 

[ban-all-sheds]

That interwebby thing must be full of info on how they work?