Does Dedicated Circuit Monitoring Help Improve Native Detection of Other Devices?

It seems that dealing with the superimposed power signatures from many devices would be one of the main challenges on the device detection front. In other words, I assume the algorithm would more detect devices more quickly and accurately in a home with 5 total electronic devices, than it would in a home with 40+ electronic devices. It seems logical that it would be easier to pick one from the crowd if the crowd itself was smaller. When DCM is used on a device (or devices if using DCM on a branch circuit), in theory the signal from the flex sensors could be subtracted from the main signal that native device detection is attempting to be performed on.

Based on this reasoning, does dedicated circuit monitoring contribute any improvement to the native device detection being done on the mains? I understand the amount of benefit DCM could provide would vary based on the device(s) on the DCM circuit and the main CT clamps, and in some cases could offer no tangible benefit at all. To state my question another way, are the signals from the DCM CT clamps removed from the main CT clamp signals before being passed to the native device detection algorithm? If this is the case, and my assumption that native detection is easier with a lower quantity of superimposed devices, then it seems that in theory DCM could offer some benefit to how quickly and/or accurately devices are detected (including improved On/Off detection).

I recall a question being asked previously on why device detection couldn’t be performed on a circuit of multiple devices connected through the same smart plug with energy monitoring. It’s understandable that the type and quality of the data from a smart plug is not even close to the complex signals from the CT clamps (although it does feel like some value could be extracted from the smart plugs in terms of signal separation on the mains). But in the case of DCM, the nature and quality of the signal from the flex CT clamps should be the same as the main CT clamps. And therefore should allow the DCM signal to be directly subtracted from the main signal.

Justin had clarified in this linked post that native device detection is not currently performed on DCM of a branch circuit. If there is benefit to reducing the number of individual device waveforms detection is being performed on, it may be more desirable to use DCM on a branch circuit to remove a higher # of devices from the main signal. For this reason it would be nice if native detection could also be performed on a DCM branch circuit.

Good questions @matthew_lasorsa. I spoke with @MaheshAtSense this morning, who provided some clarity into your main questions.

  1. Dedicated Circuit Monitoring does not contribute directly to the user’s device detection. That is, if there are say, 5 devices on the DCM circuit, those 5 devices do not necessarily get better models. Since we don’t subtract the signal, it doesn’t mean that the rest of the house will be quieter.
    However, we are actively using DCM to improve our modeling machine learning algorithms.

  2. Signals are not subtracted. To avoid duplication, if the user merges an existing device with DCM, then we will display the DCM wattage, instead of the model they merged.


Sense’s simple, intuitive real-time Power Meter, and the Power Meter views for devices (native detection, smartplug and DCM), hide a lot of underlying complexity. When we see the Power Meter waveform, we’re not seeing what is really happening to the 60 cycle per second current and voltage waveforms (and fast sampling) that Sense combines to deliver the 1/2 second updates in the Power Meter. Not to go into the math, but one can’t just subtract off one waveform from another to “reduce the noise” because AC voltage and current are phasors (they have amplitude, phase, and frequency), and power is a function of the amplitudes and phase relationship between the two. Subtracting a DCM power waveform from the Power Meter would actually distort the transition info Sense uses to do native detections.

1 Like