LTC3454 Performance Measurements

[BillyNoMates]

I've been spending some time building a simple datalogger to take multiple voltage and current measurements with the aim to analyse the performance of various LED driver designs.

Now that I've collected my first set of results, I thought I'd post them here in case anyone was interested.

Points to note:
1) LED drive current set to 395mA (according to calculation in data sheet)
2) Measured LED current ~370mA (gives an idea of the accuracy of the measurements)
3) Low Vf of LED (2.95V) - Luxeon REBEL, neutral white
4) Quoted capacity of battery: 650mAH
5) Measured capacity of battery: 637mAH
6) Average efficiency of driver over the full run = 84%

Graphs:

 

[^Gurthang]

Nice work for the first try. Keep up the good work it looks like it could be very useful.

^G

 

[Mr Happy]

Nice.

For a couple of suggestions: you might like to apply a filter in the graphing software to smooth out the noise, and you could plot the sense voltage on a different scale so that the value is easier to see?

How did you capture the data and produce the plots? You didn't say in your post. For many people the datalogger will be as interesting as the LED and driver performance. :thumbsup:

 

[BillyNoMates]

Nice.

For a couple of suggestions: you might like to apply a filter in the graphing software to smooth out the noise, and you could plot the sense voltage on a different scale so that the value is easier to see?

How did you capture the data and produce the plots? You didn't say in your post. For many people the datalogger will be as interesting as the LED and driver performance. :thumbsup:

Thanks for the suggestions. Yes, a filter would beneficial - especially on the current measurements. I also intend to take more care over the flying wires etc to reduce the noise pickup in the first place. The plotting of the sense voltage was really just a curiosity - I'll probably not bother including that and just produce a single mean value for the whole test - it should just vary with LED drive current.

The data logger was based using a PIC microcontroller demo board. The CPU has a 10-bit ADC built in which can be multiplexed through 9 different analogue channels. I'm only using 6 channels at the moment, plus one to measure the internal reference of the PIC.
Measurements are taken every half second, each of the ADC readings are taken in sequence (it takes about 5us to make a single measurement, so all the channels are sampled within 35us). Next, the stored ADC values are converted to decimal and squirted out the serial port in comma delimited format. A PC can then act as the logger, using any terminal program (hyperterminal). The file can then be loaded into the plotting program (excel or whatever).

There are some improvements that need to be made. I intend to introduce a separate supply for the logger - currently this uses the same supply as the driver and so I can't measure VLed higher than Vcc (ie boost mode).

 

[MorePower]

Any chance you'd be willing to build and sell a few of those drivers? They'd be perfect for a 3C Maglite build with a single Rebel or Seoul P4.

What was the approx cost of components for the driver?

 

[Tohuwabohu]

Thanks for the suggestions. Yes, a filter would beneficial - especially on the current measurements. I also intend to take more care over the flying wires etc to reduce the noise pickup in the first place.
...
The data logger was based using a PIC microcontroller demo board. The CPU has a 10-bit ADC built in which can be multiplexed through 9 different analogue channels. I'm only using 6 channels at the moment, plus one to measure the internal reference of the PIC.
Measurements are taken every half second, each of the ADC readings are taken in sequence (it takes about 5us to make a single measurement, so all the channels are sampled within 35us).

A simple RC lowpass at each converter input could help to reduce noise.
And you could do some averaging in the PIC before sending the data over the serial interface. Instead of taking only one sample per channel you could take several, add the readings and divide the sum by the number of samples. A very easy way would be to add 64 readings and then shift the sum 6 bits to the right to get an averaged 10 bit result.

Next, the stored ADC values are converted to decimal and squirted out the serial port in comma delimited format. A PC can then act as the logger, using any terminal program (hyperterminal). The file can then be loaded into the plotting program (excel or whatever).

Do you know LogView?
You could send your data in the 'OpenFormat' and use Logview to visualize the data while the test is running.
Unfortunately most of the documentation is only available in german.

 

[HarryN]

Hi - nice graphs from that data logger project. I had wondered when you would start posting about it.

It is unclear to me if the signals are that noisy, or possibly the sources really are pulsing around like that. (ripple, pwn, etc). I sell spectroscopy equipment, and I am always amazed at how much additional information an expert can be pull from what I first interpret as "noise". Data averaging is handy, but not always realistic.

In any event, it does look like a very handy tool. Hopefully at some point I will be ready to try one.

 

[BillyNoMates]

Any chance you'd be willing to build and sell a few of those drivers? They'd be perfect for a 3C Maglite build with a single Rebel or Seoul P4.

What was the approx cost of components for the driver?

I could potentially offer a some drivers for sale, but that wasn't my main intention for this project. I tend to bundle multiple drivers and an controlling PIC onto a single board and that is not a popular solution for many modders.

I could drop a single driver option onto the next PCB panel. Any advice on what PCB diameter, whether it is single sided or not, heatsinking preference etc etc would be helpful.

I'm not a big manufacturer, so component cost is not as low as it could be - I'm looking at a typical cost of £4 per driver, excluding PCB.

 

[BillyNoMates]

A simple RC lowpass at each converter input could help to reduce noise.
And you could do some averaging in the PIC before sending the data over the serial interface.

Thanks for the suggestion, both of those are on the to-do list . I'll probably just start with 16 samples as that should reduce the output variance enough.

Do you know LogView?
You could send your data in the 'OpenFormat' and use Logview to visualize the data while the test is running.
Unfortunately most of the documentation is only available in german.

I've not used LogView before - I may give that a try when I next play with this.

 

[BillyNoMates]

It is unclear to me if the signals are that noisy, or possibly the sources really are pulsing around like that. (ripple, pwn, etc). I sell spectroscopy equipment, and I am always amazed at how much additional information an expert can be pull from what I first interpret as "noise". Data averaging is handy, but not always realistic.

There is always a lot going on in the switching action of the converter, but that information is contained at much higher frequencies than the sampling rate I'm using here. If I sample twice per second, then I can't capture any information with a frequency content of more than 1Hz (Nyquist rule). Any signals with a frequency content greater than this are 'aliased' and can't be reconstructed and behave like noise/interference. If I was being rigourous I would need to put a 1Hz filter onto the sample channels (I haven't done this). Alternatively, I use a wider filter and sample more frequently (as suggested by Tohuwabohu).