Energy Fidelity Scalability

Advancements in technology are pushing the bounds of on-body sensing, signal processing, and communications, resulting in a host of new and useful platforms with compelling applications. Body-area sensor networks (BASNs) show particular promise for improving the medical field by allowing for information to be gathered unobtrusively in naturalistic settings – improving care and quality of life while reducing costs. As a result, the design of BASNs is now guided by many of the same requirements found in consumer mobile electronics: package miniaturization, longer battery life, and intelligent and sophisticated functionality. However, BASN efficacy is also predicated by application-oriented, outcome-focused demands, which are strict requirements in a number of BASN applications, especially those with a healthcare or medical research focus. Sustaining the level of innovation necessary for current and future BASNs to achieve these competing aims requires new methods, tools, and frameworks to evaluate and manage the complex, application-centric tradeoffs that system designers now face.

Since a large amount of the energy consumed by a BASN node is due to the wireless radio used to stream data to an aggregator on or off body, significant power reduction can be achieved through the development of scalable, on-node techniques like signal processing and data management, thus dramatically reducing the number of bits to be transmitted. Low power signal processing therefore becomes increasingly important to BASN power efficiency. Data rate reduction often has a positive impact on energy consumption, and consequently wearability and run-time, but a negative impact on guaranteed fidelity. Other techniques can also be used alongside embedded signal processing to increase energy efficiency and system run-time. The embedded tools available on BASN nodes that allow for exploration of the relationship between data reduction and signal fidelity can be put into two categories: inputs (measurement and analysis) and outputs (“knobs” to turn). The relationship between data reduction/compression rate and resulting fidelity must be understood in order to optimally trade-off these two competing metrics.
Previous work [1] by the INERTIA team has shown the existence of an energy-fidelity tradeoff in BASNs with digital signal processing employed. This research used Haar wavelet compression and rate-resolution scaling as example lossy data reduction schemes for use in exploring the tradeoff space since they met the following three criteria:

  • capable of being implemented on resource-constrained BASN embedded processors,
  • capable of executing in low-latency and soft real-time applications, and
  • adjustable by key “knobs” to alter expected data reduction rates.

Mean Squared Error (MSE) was used to assess fidelity as is commonly done in the signal processing community, and the results indicated there is a large energy-fidelity exploration space possible in BASNs. The figure below shows a small portion of this space using the Haar wavelet transform and run length encoding for data compression and highlights another interesting fact: the input signal characteristics change the possible energy-fidelity operating points.

Moreover, it is interesting to note that the data shown in the above figure is from a single patient over the course of a single clinical visit. The amount of “information” present in the sensor signals changes over time along with the rate-distortion curve pointing to the need for dynamic management of energy-fidelity tradeoffs in these embedded environments. To illustrate further, the figure below depicts a time domain distortion plot for fixed data compression, yielding a compression ratio (CR) of approximately 18, for a 40 minute tremor dataset. Thus, merely choosing a static operating point on a curve of the above figure is not sufficient for application fidelity regulation or energy efficiency.

In fact, our recent work [2] has shown that even simple schemes using dynamic compression outperform optimized static settings. Specifically, this work showed that Haar Wavelet compression can be dynamically varied based on measuring the incoming variance to operate at an energy point not possible by statically setting the Haar compression ratio (shown in the figure below).  In addition, DVFS techniques can be used on-top of this approach to further improve the energy-fidelity trade-off.

Future BASN devices must therefore possess energy awareness (knowledge of how much energy has been consumed), data awareness (knowledge of how compression affects current data), and computing resource awareness (knowledge of how algorithm execution affects processing and memory resources) to effectively tradeoff runtime and output fidelity in a way that is executable on resource constrained platforms and that meets real-time requirements. These tradeoff decisions can be made based on efficiently meeting requirements (e.g. maximum lifetime for a given minimum fidelity, maximum fidelity for a given minimum lifetime, etc.) or minimizing bounded cost functions (e.g. minimizing lifetime-α•fidelity-β given minimum lifetime and fidelity requirements, where α and β are determined based on metric priorities).  For instance In applications like fall prevention, gait analysis, or tremor monitoring, the system may need to collect data for multiple days, but the activity level of the subject is unpredictable.  The system should adjust the energy-fidelity operating point to provide a personalized solution and deliver the best data for the desired time period.

This work uses embedded hardware and software techniques to dynamically manage the energy-fidelity relationship that exists in many BASN applications thereby enabling smaller, more wearable devices with longer runtimes. This research will allow individual BASN nodes to intelligently and quickly respond to changing sensed data characteristics. Limb-worn gait and tremor inertial patient data collected from the TEMPO (Technology Enabled Medical Precision Observation) system is used throughout this research as an example of small form factor, high data rate BASN applications. A system architecture is being designed that allows for dynamic calibration of data processing and transmission/storage to adjust for variable incoming data rates and dynamic wireless channel characteristics. The architecture shown below illustrates the use of both a data controller and a destination controller for runtime optimization of tradeoffs between power consumption, computational complexity, and signal fidelity. This system architecture is intended for implementation on resource constrained platforms common to BASN nodes.

The “destination controller” attempts to mitigate variations in the wireless channel and to ensure reliable data transport in the unpredictable body-worn environment. The current research however focuses on the design of the “data controller” concerned with manipulating the processing of incoming sensor data to ultimately reduce the amount of information sent over the wireless channel. The controller must adjust to varying incoming data characteristics since BASN sensor signals are often variable and unpredictable.
In general, this research attempts to understand the energy-fidelity tradeoffs present in modern day BASNs and strives to answer the question: “How should system designers and application experts make decisions about what energy-fidelity management schemes are appropriate for their application?”  A generalized software toolchain was designed to investigate energy-fidelity tradeoffs in the context of various applications which allows a system designer to input their own BSN hardware information and previously collected information.  The toolchain was written in LabVIEW.  The source code can be used to input new parameters and can be found HERE.  An installer which incorporates specific hardware energy models and processing/compression schemes can be found HERE.  Finally, documentation describing the software and where to make modifications can be found HERE.

 

Relevant Projects:

TEMPO 3.1          Dynamic Voltage-Frequency Scaling          Tremor Assessment          Hemodialysis Fall Prevention            Cerebral Palsy Gait Analysis    


References

  1. Hanson MA, Powell, Jr HC, Barth AT, Lach J.  2009.  Enabling data-centric energy-fidelity scalability in wireless body area sensor networks. Proceedings of the Fourth International Conference on Body Area Networks. :1u20138.
  2. Barth AT, Hanson MA, Powell, Jr HC, Lach J.  2010.  Online Data and Execution Profiling for Dynamic Energy-Fidelity Optimization in Body Sensor Networks. 2010 International Conference on Body Sensor Networks. :213u2013218.