benz

Measure this, not That: Signal Separation in Optical and Electrical Measurements

Blog Post created by benz on Oct 13, 2016

Originally posted Oct 12, 2015

 

I wonder if these MIT researchers can quantify their directivity?

I’m fascinated by correspondences between phenomena in RF engineering and other fields, and it isn’t just a matter of curiosity. These correspondences are also enlightening, and sometimes guide genuine technological advances.

An interesting cross-domain example is the recent MIT announcement of a technique for removing unwanted reflections from photos taken through windows. We’ve all experienced this problem, feeling the surprised disappointment when the photo includes obvious reflections we didn’t notice when composing the picture. At least with digital cameras, we can usually spot the problem while there’s still a chance to take another photo and fix or reduce it.

That surprised disappointment is actually a pointer to the kind of solution the MIT folks have produced. If you haven’t seen it already, take a look at the before/after in the MIT press release.

The uncorrected image is likely to be familiar, and the strength of the reflections is often much greater in the resulting image than it was perceived by the photographer. The perceptual shift is likely caused by our visual system’s ability to automatically do a version of what the MIT technique attempts to do: separate the reflection from the desired image and subtract or ignore it.

The MIT technique doesn’t identify the reflection directly, but it can recognize pairs of them. That’s useful because the unwanted reflections often come from both the front and rear surfaces of the intervening glass, with an apparent offset.

Unwanted reflections from photography through a window—such as the photographer’s hand or body—usually appear in offset pairs, originating from both the front and rear surfaces of the glass. Blame me, not MIT, for any errors or oversimplification in this diagram.

Unwanted reflections from photography through a window—such as the photographer’s hand or body—usually appear in offset pairs, originating from both the front and rear surfaces of the glass. Blame me, not MIT, for any errors or oversimplification in this diagram.

When reading about the technique, my first thought was the similarity to network analysis and its powerful tools for separating and quantifying incident and reflected energy. The analogy breaks down when considering the separation methods, however. The gang at MIT look for the reflection pairs, perhaps with something similar to two-dimensional autocorrelation. RF/microwave engineers usually make use of a directional coupler or bridge.

Directional couplers separate incident and reflected energy, and a critical performance parameter is directivity or how well the coupler can separate the energy moving in each direction.

Directional couplers separate incident and reflected energy, and a critical performance parameter is directivity or how well the coupler can separate the energy moving in each direction.

Of course, I now find myself wondering about the effective directivity of the MIT separation-and-removal scheme, and if they think of it in those terms. Probably not, though that would be a ready-made way to quantify how well they’re doing and it might help in optimizing the technique.

Recently, I’ve written about improving measurement accuracy. However, in thinking about these tools and techniques, I realized that separating signals to measure the right one is fundamental to making better RF measurements of all kinds. Indeed, the separation process is often more difficult than the core measurement itself.

Spectrum analyzers naturally use their RBW filters to separate signals into their different frequency elements, but it may also be critical to separate them by their behavior or their time duration and timing, or to separate them from the analyzer’s own noise.

 

I could go on and on, and branch off into optical separation techniques such as steganography. Now that I’m looking for such methods, I see them everywhere and resolve to consider signal separation explicitly as an essential step to accurate measurements.

Outcomes