IOT Power Profiler and battery life estimation

[EEEnthusiast]

Hello Friends,
I have been trying to solve the current measurement problem faced by many IOT developers. The problem is with high dynamic range and high bandwidth measurement. The current drawn by IOT devices can change from few uA to 100s of mA in couple of micro seconds. A typical profile cycle for an IOT device can last several minutes, if not hours. Also most of the developers need to estimate how long their batteries would last for a given design and power profile. I was able to address all of these in a new product design.

The tool is called IOT Power Profiler with the product number ZS-2102-A. It is currently on sale for $599 on the website www.anglercircuits.com which also provides descriptive video and specification sheets along with some other useful information.

The ZS2102A can measure currents from few uA to 1A in both directions. It has an accuracy of 1% +/-1uA with < 100nA resolution. It can accept batteries up to 6V range and can record power profiles for longer than 24 hours at a stretch. It uses constant resistance of 0.1Ohm across the range and uses some precision analog components to achieve this high dynamic range. The samples are streamed over the USB 2.0 interface and compressed on the fly by the GUI running on the PC. A 24 hour session could be recorded on a 10GB file. It then uses real world battery models to estimate the battery life using the current profile.

There are couple of power profilers out there already. With due respects to all of them. this one is an addition which tries to fix some of the gaps I saw in them. Your comments and feedback on this is highly appreciated.

Thanks.

[EEEnthusiast]

The article was read more than 100 times but not even a single comment.
Either the tool is just too good to find fault, or just not fitting the bill at all.
I would really appreciate if some one points out some short fall of the equipment so that we can make necessary changes to the hardware or software.

By the way we have released a new version (Ver 2.1) of the software which enables data analysis (viewing the profile) and measurements at the same time. This came in as a feedback from the user where he is continuously measuring current profiles after changes to the firmware and does not want to wait till the capture is completed. The new release also reduces the CPU loading to about 25% on a Core i3 @2.3GHz. The RAM usage is reduced to 512MB and below at all times. So the tool can run on almost all Windows machines.

Next-up we would release a viewer only program which can run without the tool for the data to be analyzed on any computer.

Many are suspicious about the tool achieving high dynamic range from few uA to 1A and accuracy of 1%+/-1uA across the range with just a 0.1 Ohm burden resistor. This is possible only with some precision analog components and some wacky hardware design along with signal processing on the GUI.

The hardware does not perform any signal processing, and this makes the equipment noise very low. The tool just streams raw samples to the PC using a FT2232H IC. The GUI does the all the DSP operations and data compression. The compression achieved is almost 100x. For e.g. a 24hour recording is possible on a 10GB file and this makes it possible to record long profiles.

I was working as an application engineer for Texas Instruments for 17 years and was experimenting with a lot of precision circuits over there.

[EEEnthusiast]

Thanks BlueSkull,

I can post some internal pics of the tool. No secrets there. After a few years, I plan to have the hardware files online so that anyone can assemble it with ease.
Sure I could send one to Dave. I had sent him multiple emails for an address to send the product , but have not yet received a response. May be this thread will kindle a response from Dave.. 

[EEEnthusiast]

Ready for that...

[windsmurf]

Can this be used like a current-measurement oscilloscope?

What would you say is its advantages and disadvantages over something like a mooshimeter?

[EEEnthusiast]

Yes, it is a current oscilloscope albeit a precision one with high dynamic range. The Mooshimeter is a precision one with high dynamic range, but it does not have the bandwidth needed for measuring short current spikes which are few tens of micro seconds wide.  The ZS2102A combines the response of a scope with the precision and dynamic range of the mooshimeter/multimeter into one unit. That is what makes it unique. In terms of function, the closest match is the JS110 Joulescope.

Mooshimeter cannot respond to a current pulse which is say 50us wide and rises from 100uA to 100mA in 2us. But the ZS2102A can accurately measure and plot them. This is significant for measurement of battery life of IOT systems as most of the current profile consists of such short spikes.

[nctnico]

The article was read more than 100 times but not even a single comment.
Either the tool is just too good to find fault, or just not fitting the bill at all.

It would have been nice that you where clear about the fact that this is your product and your post is kinda like an advertisement...

More on-topic: what I'm wondering about is how you achieve the large dynamic range. I guess you might be using a lin-log converter. I have been thinking about a product which is much simpler for use with an oscilloscope. When using a lin-log converter the oscilloscope can use math to transfer the captured waveform back to a linear current reading if necessary.

[EEEnthusiast]

My bad... the intention was more to brainstorm on the product and get some feedback..

Coming to the technical point, the dynamic range is increased by dividing the entire range into blocks and using different gains in each block. For e.g. in the uA range, the gain is higher than in the mA range. This improves the resolution. To achieve low drift, there are chopper stabilized amps and dynamic offset cancellation. The bandwidths in each block is set to keep the noise below a certain level to get the desired accuracy. Then the measurements from the different ranges are combined on the PC to get the overall total measurement.

[nctnico]

That is a simple solution but it would mean dealing with overdrive recovery effects in the amplifiers as well when the signal makes a large jump (besides the 'weirdness' some amplifiers exhibit at the input when overdriven).

[EEEnthusiast]

The observation about the overdrive is very much correct. But the overdrive effect is already taken care in the design. The step response of the tool is shown below. This is measured from 0 to about 0.5A. The step response is clean and has no effect of amp saturation. Similarly on the way down, there are no overdrive recovery artefacts seen. (the pic with the GUI snapshot) I can post some pictures, with slow ramp up and ramp downs to show the switchover clearly, but these are also clean.

The different ranges have overlaps which provides the op-amps enough time to recover before the range switching takes place. This is the hardware level of overdrive recovery elimination. On the software side, the data from all the ADCs are streamed to the PC and the PC GUI makes the decision of optimal combination of these in order to provide the best estimate of the current. It is not just a range selection which is handled by the GUI but a smooth blending of the data. So the overdrive effects are almost zero here.

[Mr.B]

The creator of the JS110 Joulescope is or was a member here.
Dave briefly reviewed the Joulescope here: https://www.eevblog.com/forum/blog/eevblog-1190-mailbag-review-bonanza/
I have been considering purchasing such a device.
Other than price, what would convince me to purchase the ZS-2102-A over the Joulescope?
Thank you in advance.

[EEEnthusiast]

On the ZS-2102-A here are few features which stand out among others
1. Better accuracy : 1% +/-1uA across the whole range from -1A to 1A

2. Long capture time due to data compression. You can capture few hours of data at a stretch ( I have tested it for 24 hours straight and the data is no bigger than 10GB for a WiFi sensor)

3. Constant resistance : This implies that there are no artefacts of resistor switching present in the current measurement. Switched resistors may exhibit noise, when the current is just at the threshold of the switching, and when there is extra noise overriding the current. This can change the average value, as the noise caused by the switching may not be having Zero Mean value.

4. Battery models : The GUI comes with real battery models which can be used to estimate the battery life. This is not just dividing the mAH number by the average current. The GUI analyses the current profile sample by sample, and along with the battery model to estimate the battery life precisely. For e.g. a profile with higher peak current may cause batteries to deplete faster as the chemistry gets altered during high spikes.

5. No processor or FPGA on the hardware. There is only a CPLD and FT2232H along with a bunch of high precision analog components. This calls for a lower noise design, free of spurious emissions which can be caused by the processors/fpga. All the updates are on the GUI which is quite easy to upgrade. Just download and install the software on the PC. No need to update the fpga or processor firmware.

6. Better export features on the GUI. The GUI can export to Sigrok format, which can be viewed on PulseView  to analyze the current waveform together with the digital IO, which can be analyzed for various protocols. For e.g. one can check for messages on the UART coming out of the DUT while analyzing the current. This is useful for debugging firmware and software.

7. Better error rate : We have captured data for extended times (> 24 hours and multiple units and multiple iterations) without a single error. The estimated error rate is < 1E-12 which is quite difficult to achieve. This implies there are no sparkle codes from the ADC, no glitches and USB transaction errors.

8. Uses default drivers installed on Windows for FTDI. Ensures better compatibility and stability.

[EEEnthusiast]

At the same time, I would like to point out some shortcomings of the tool.

1. The voltage input is sampled only at 20ms interval. The assumption is that the battery voltage will not drop drastically over short intervals. However high and fast current spikes can drop the voltage down in few us. The tool will miss those spikes. However these voltage drop could be calculated to some extent on the GUI as the current information is available. This is however not performed on the GUI.

2. The terminals are not isolated from the ground on the USB interface. This could cause problems if the user is using another grounded power supply, instead of a battery to power the DUT. Ground loops can pick up noise and may cause accuracy issues.

3. The voltage at the input is limited to 9V. The system is designed for IOT devices, which are mostly powered by Lithium or Li-Poly batteries, which reach a maximum of 4.2V. 9V is kept as a limit for some systems which use a 9V alkaline battery like door bell.

4. The software supports only Windows (Win7 to Win10). There is no Linux or Mac support available today. But these are in the pipeline and expected to be available by the end of 2019.

5. The current is limited to 1A for accurate measurement. The tool can accept up to 3A for short pulses.

[nctnico]

No USB isolation can be a deal breaker. Why didn't you make the measurement circuit isolated? For starters it could help with noise.

[EEEnthusiast]

The tool is intended to be used with a battery powered system. Adding isolation on the analog section was causing non linearity issues and on the digital side these would have caused speed issues as most opto-isolators are slow. There are some high speed isolators like ISO721M-EP but I had seen some issues in using them. Then there was another issue of power delivery with isolation and that would have needed to add some DC-DC converter which is isolated adding noise and space.

In the ZS2102A, there is still some isolation (few KOhms) but still not like the opto-isolators.
Even without the isolation, the tool can be used to measure the currents of loads powered by regular bench power supplies or USB powered devices. The noise is seen only in some cases and it is not very severe. I can post some measurements once I'm back in the lab showing the comparison.

A simpler approach I found was to run the GUI and the tool from a laptop, which is disconnected from the mains supply. This way the tool gets isolated from the mains ground. Thus the ground loop is broken and the ground noise gets reduced.

[mliberty]

Disclaimer: I am the creator of Joulescope

Here is the link to the Joulescope website where you can download the Joulescope User's Guide including specifications.  Congratulations to @EEEnthusiast for creating a product that also addresses this very real problem.  I take issue with a number of the ZS-2102-A "features which stand out", but I'll leave it to the community and reviewers to make the comparisons.

[EEEnthusiast]

Greetings MLiberty,
I have read through the JS110 specs. Needless to say that the tool has better dynamic range than the ZS2102A (can measure few 100nA to 10A accurately). It has similar bandwidth as ZS2102A but uses 2x sampling rate. It can accept higher voltages and has lower series drop at steady state. The outputs are isolated from the USB ground. Great product I would say honestly as an engineer. We have both started off with the same problem and converged on two different solutions. Both has merits and demerits.

I had thought through similar lines to use a MOSFET switched resistor shunt, but could not solve the range hunting problem reliably. If that issue is fixed in the JS110, then I would buy one to evaluate.

The problem is when the input current is at the threshold of switching (Let us say 18mA) and there is a current ripple on top of that of 1mA. A very slow ramp response or triangular wave response of the tool can help locate this issue. Step response will miss this, due to the fast transition.

I had also faced the issue of gate/drain leakage current causing measurement error at low current values. The leakage of many mosfets can be several uA and this does not pass through the burden resistor but will pass through the load. This will contribute to some error which has a high temp coefficient. So calibrating this is a difficult task, if not impossible.

With due respects to MLiberty for solving a complex problem, I think the JS110 is a great product to buy.

[Mr.B]

Thank you for your explanation @EEEnthusiast.
Unfortunately the lack of isolation will mean your device is not suitable for my needs.
While I intend to use the device for IoT projects, I have other test scenarios I wish to cover off as well.
I wish you well with the sales of your device.

[coppice]

Greetings MLiberty,
I have read through the JS110 specs. Needless to say that the tool has better dynamic range than the ZS2102A (can measure few 100nA to 10A accurately). It has similar bandwidth as ZS2102A but uses 2x sampling rate. It can accept higher voltages and has lower series drop at steady state. The outputs are isolated from the USB ground. Great product I would say honestly as an engineer. We have both started off with the same problem and converged on two different solutions. Both has merits and demerits.

The JS110 has serious drawbacks. It uses range switching to cover its wide dynamic range, and incurs serious performance issues as a result. If your solution genuinely covers the wide range of currents in a single range, with no dead time and bandwidth issues as the current hops over that wide range, you seem to have a most effective design.

[EEEnthusiast]

Greetings MLiberty,
I have read through the JS110 specs. Needless to say that the tool has better dynamic range than the ZS2102A (can measure few 100nA to 10A accurately). It has similar bandwidth as ZS2102A but uses 2x sampling rate. It can accept higher voltages and has lower series drop at steady state. The outputs are isolated from the USB ground. Great product I would say honestly as an engineer. We have both started off with the same problem and converged on two different solutions. Both has merits and demerits.

The JS110 has serious drawbacks. It uses range switching to cover its wide dynamic range, and incurs serious performance issues as a result. If your solution genuinely covers the wide range of currents in a single range, with no dead time and bandwidth issues as the current hops over that wide range, you seem to have a most effective design.

The ZS2102A covers the range from few uA to 1A in both directions with a single resistor at all times while maintaining the 1%+/-1uA error.  There is no range switching and no dead times at all.  There are multiple gain stages after the 0.1 Ohm sense resistor to scale up & improve resolution , accuracy and some stages to cancel drift. The data from multiple ADCs is combined optimally using some DSP algorithm on the GUI and hence the blending is smooth. The hardware just measures and dumps the data over the USB. No intelligence is built into the hardware and that makes it simple. It only uses a bunch of high precision OPAMPs , references, reference buffers and ADC with precision ADC drivers. Almost 90% of the BOM cost is analog precision components.

[EEEnthusiast]

Apart from the ZS2102A hardware, the GUI plays an import function of data compression. Without the compression, the raw data rate is about 8Mbyte/sec and this can cause problems with other software running on the PC. If some other program writes to the disk with continuous data access for a long time, then the data capture tool will not get enough time to write the samples. This can cause buffer overrun and data errors.
In case of the ZS2102A, the data is compressed by a ratio of more than 100x. Thus it can store the data in the PC memory for a long duration (several minutes) before writing to the disk. The final data size is roughly 500MB per hour which is quite reasonable, given the disk sizes we have today. This enables the ZS2102A to capture long profiles of several hours, if not days.

[EEEnthusiast]

Friends,
I have added another topic for a new design for a IOT Power Supply. Still in  R&D phase with everything on paper.
https://www.eevblog.com/forum/testgear/iot-power-supply-design/

[EEEnthusiast]

Friends,
I am looking to manufacture the IOT Power Profiler tool in the USA due to high import duty in India, high shipping costs etc.
If you can provide me leads to few companies in the USA, who can take up the whole turn key manufacturing and/or sales and marketing, it would be very helpful.
Thank you

[Kean]

Friends,
I am looking to manufacture the IOT Power Profiler tool in the USA due to high import duty in India, high shipping costs etc.
If you can provide me leads to few companies in the USA, who can take up the whole turn key manufacturing and/or sales and marketing, it would be very helpful.
Thank you

This is a useful reference: https://www.crowdsupply.com/providers

[EEEnthusiast]

Kean
Thanks for the link. That is super useful.

Do you happen to know any company, who can take up entire task of manufacturing and sales together by paying a royalty or revenue share. The logistics of manufacturing , sales and marketing is just time consuming and energy draining, without major electrical engineering challenge.