Compressive multispectral sensing algorithm with tunable quantum dots-in-a-well infrared photodetectors

In recent years, our group has developed and reported two multispectral sensing algorithms that aim to exploit the continuous bias-dependent spectral tunability of the quantum dots-in-a-well (DWELL) infrared photodetector and enable higher spectral resolutions without using spectral filters. The key idea is to probe an unknown target of interest sequentially with the DWELL detector at multiple biases, producing a set of bias-dependent photocurrents. Then, a post-processing algorithm performs a linear superposition of these bias-dependent photocurrents with a pre-determined set of weights, which is the optimal solution for a specific multispectral sensing task. The first algorithm, termed the spectral-tuning algorithm, is designed to perform algorithmic spectrometer, which is capable of reconstructing the spectrum of any unknown target of interest (admitted by the DWELL's spectral response) without utilizing any physical spectral filters or spectrometer. The set of weights obtained by the spectral-tuning algorithm can optimally approximate the desired shape of narrowband tuning filter with a specified bandwidth. According to this optimal set of weights, the spectrum of the unknown target at each desired tuning wavelength is reconstructed by performing a weighted linear superposition with the set of bias-dependent photocurrents. This algorithm has been experimentally demonstrated by our group [1] and other groups [2]. The second algorithm, termed, the spectral matched-filtering algorithm, is geared toward performing target classification [3]. With known multiple spectra, representing classes of targets of interest, the idea is to obtain the optimal sets of weights, which form a linear superposition with the bias-dependent DWELL's spectral bands. Each superposition band is regarded as the most informative "spectral direction" (in a vector-space sense) for a given target spectrum. Then, the classifier uses these sets of weights to perform a linear superposition with the measured set of bias-dependent photocurrents. The outcome from the classifier is the set of extracted superposition features, which is used to classify the unknown target.