Digital signal processing (DSP) is a skillset that enables analyzing and extracting useful information from digitized sensor data collected in the environment. Examples of signal sources are sound, light intensity, light color, vibration, electric field, magnetic field, electrical impedance, temperature, humidity, volatile organic compounds (VOC), CO2, CO, flexure, acceleration, and angular rotation rate. These variables can be used alone or in combination to estimate useful information like air quality, occupant comfort, dew point, device orientation, heart rate, electrodermal activity, and many more applications. Often, measured signals contain a significant amount of noise and the goal of DSP is to extract the useful signal from amongst the noise.
Process
In some cases, a problem will be straightforward enough that an experienced engineer can identify and implement a working algorithm that satisfies all requirements fairly quickly. An example is using two-point calibration to convert resistance of a thermistor to temperature.
However, many estimation problems are not readily solved by a one-size-fits-all algorithm. The best way to tackle a complex estimation problem is to use a divide-and-conquer approach. First, use separable features to classify the problem into two or more categories. Second, apply a targeted algorithm for each specific category. An example of this approach is when estimating physiological signals in fitness oriented devices. Often the optimal estimation algorithm structure or parameters will vary depending on the context of what the user is doing. For instance, signals while the user is sleeping are vastly different than when that person is weight lifting. We apply a different type of algorithm for each of these domains.
To execute this successfully requires skilled application of classification techniques such as machine learning (ML), as well as a breadth and depth of knowledge in applying estimation algorithms such as Kalman Filtering, regression, frequency analysis, etc.. DSP is a challenging but rewarding puzzle-solving endeavor which benefits greatly from experience looking at signals and going through the trial-and-error process. Experienced DSP engineers will be able to look at a noisy signal and intuitively know which algorithms work the best, not wasting time trying dead-ends. Therefore, working with an expert can save significant time and money even if their salary or rate is at a premium. If you have a challenging DSP problem that could use an expert eye towards it, please reach out and contact us.