Energy Efficient Compression of Shock Data
Using Compressed Sensing
Jerrin Thomas Panachakel and K.C. Finitha
Abstract This work analyses the potential of compressed sensing (CS) for com-
pressing shock data signals of space launch vehicles. Multiple shock data signals
were compressed using compressed sensing by exploiting the sparsity of the shock
data signals in the time domain. Since shock data signals are sparse in wavelet do-
main also, thresholding based DWT compression was performed to compare the
performance of compressed sensing. Three performance metrics, viz. Peak Root
mean-square Difference (PRD), Compression Ratio (CR) and execution time were
used. It is also evaluated how compression of the shock data reflects in the Shock
Response Spectrum (SRS). The results clearly show that CS surpasses the DWT
based compression in terms of execution time for a given C R but has slightly in-
ferior results in terms of P R D for higher values of C R. With lower computation
power requirements and dimensionality reduction, CS becomes an ideal choice for
compressing shock data signals in a mobile signal processing system with constraints
on processing power and for transmission over a power-hungry wireless network.
1 Introduction
International Data Corporation (IDC) forecasts that the “digital universe” will grow
to an astonishing 8 ZB by the year 2015 [1]. This figure was just 2.7 ZB in the
year 2012 [2][3]. This unprecedented rise in the data creation rate calls on, never
like before, the need of compression techniques. The art and science of compression
itself has evolved by time. The first attempt to compress electrical signals was by
J.T. Panachakel(B)
Indian Institute of Science, Bangalore, India
e-mail: [email protected]
K.C. Finitha
Vikram Sarabhai Space Centre (VSSC), Trivandrum, India
e-mail: [email protected]
© Springer International Publishing Switzerland 2016 273
S. Berretti et al. (eds.), Intelligent Systems Technologies and Applications,
Advances in Intelligent Systems and Computing 384,
DOI: 10.1007/978-3-319-23036-8_23
274 J.T. Panachakel and K.C. Finitha
Homer Dudley in the late 1930s where he used what is known as the “vocoder”
[4][5]. Depending on the requirements, different methods are used for compressing
different classes of signals.
In this paper, we propose the use of Compressed Sensing for compressing the
telemetry signals transmitted from a space launch vehicle. Out of the various teleme-
try signals viz. shock data, vibration data and the acoustic data, compression of shock
data is discussed in this paper. Shock data is defined as the plot of the magnitudes of
shock pulses as a function of time where shock pulses are caused by the non-periodic
excitations characterised by severity and suddenness and which usually causes rel-
ative displacement in the system [6]. As far as a space launch vehicle is concerned,
these pulses result from rocket motor ignition, staging events, deployment events, etc.
[7]. Although it is essential to compress shock data since a typical flight test requires
more than ten three-axis accelerometers for generating the shock data [8], not many
works have been carried out in compressing shock data mainly because [9] [5],
• Typical data compression algorithms are complex and are often difficult to main-
tain.
• Typical data compression algorithms may affect the inherent structure of the data
making the analysis of the signal difficult.
• Computational power constraints in a mobile signal processing system.
The proposed CS based compression technique addresses these constraints effi-
ciently. CS was our choice of compression scheme because,
• Shock data signal, as discussed in Section 2.1, is sparse in multiple domains making
it an ideal candidate for compressed sensing.
• Compressed sensing has comparatively lower power requirements [10].
We believe that the reduction in the computational complexity manifests into energy
efficiency.
The rest of the paper is organised as follows, Section 2.1 discusses about shock
data signals, Section 3 covers, rather briefly, compressed sensing. The various results
which we have obtained are given in Section 4.
Notation: Normal letters designate scalars, boldface letters designate vectors and
capital boldface letters designate matrices. Also, x, y stands for the inner product
of x & y and ||.|| p for the l p norm.
2 Shock Data Signal Analysis
2.1 Saturation Analysis
Shock data signals sampled at the rate of 6400H z were used in this work. First
500 samples of one of the shock data signals used in this work is shown in Fig.
1. A common problem associated with shock data signals is the saturation of the
accelerometer used [7]. Tom Irvine in [7] notes that the positive and negative spectral
curves in a saturated accelerometer data diverges from each other in acceleration
