Analogue implementation of motion
estimators for digital video encoding
Image compression is a vital technology in the realisation
of the information highway and motion estimation (ME) is a key aspect of image
compression systems. Although simple in principle, real-time ME processors are
notoriously difficult and expensive to realise
because of their huge computational requirements. As a result, the ME processor
is the most complex and power hungry component of a digital video codec. This
project is concerned with the application of analogue circuits to ease the
implementation complexity of ME processors for digital video encoding. Using a
new architecture for ME processor design and current-mode analogue circuits,
large reductions in size, power dissipation and implementation cost can be
achieved compared to digital implementations. To demonstrate the new approach,
a compact, low power, analogue ME processor will be designed, built and
evaluated which could be readily extended to meet the requirements of
commercial digital video encoding. Such a ME processor is particularly
important in portable video systems where size and power dissipation are
incompatible with digital implementations.
Analogue
implementation of convolutional decoders using a new
decoding algorithm
Convolutional decoders have long been
important in applications where very noisy channels are encountered. The
technique has recently become more significant with the advent of new
application areas such as satellite and mobile radio systems of various types
and digital magnetic recording systems. The most commonly employed convolutional decoding technique is the Viterbi
algorithm (VA). This provides an optimum method for realising
a convolutional decoder, but requires a large amount
of digital path memory, typically up to 50% of the total chip area. This
project is concerned with the design and implementation of convolutional
decoders based on a new alternative approach called the modified feedback
decoding algorithm (MFDA). This algorithm requires no digital path memory and
so can be implemented using almost entirely analogue components, which offers
the potential for the construction of miniature low power convolutional
decoders. The MFDA approach trades implementation complexity and power
dissipation against operating speed. Such trade-offs are very important for a
number of portable applications such as emerging and future wireless
systems and low power implantable biomedical systems employing radio
communication with external sources.
An implant for
controlling neurogenic incontinence by combining neuromodulation with stimulation
We propose to design an implant system which can be used by
clinical researchers to investigate new methods for treating neurogenic incontinence. The implant will allow both ENG
recording from, and selective stimulation of the afferent and efferent pathways
of the sacral nerve roots. Our proposed implant will avoid the need to cut the
sensory nerves by using conditional neuromodulation.
In addition to being more popular with patients with complete spinal cord
lesions, successful use of neuromodulation will widen
the user group to include incomplete-lesion patients. The latter will also
benefit to the extent that pain can be prevented. All patients may benefit from
improved defecation. It is also possible that the device could benefit some
patients with other neurological conditions (eg.
Multiple Sclerosis). The two major parts of the project are the design of a
4-channel ENG amplifier integrated circuit and the design of the complete
implant system. Prototype implants will be produced.
Bi-directional
interfacing of Electronics and Cultured Neurons
The overall aim is to advance our understanding of how mammalian
nerve cells can be optimally connected to integrated electronic circuitry for
neurobiological research and medical applications. Current methods for directly
measuring neural activity use patch clamps or optical recording. Th first method is bi-directional, but does not have the
ability to make reliable one-to-one connections to single neurones
over an extended period as it is intrusive. The second method allows only
recording and requires imaging technology if an automatic system is to be used.
This implies a large amc of storage and the development
of software to interpret the firing patterns. Like patch clamping, the
fluorescent dyes used for optical imaging are not suitable for long-term
measurements: the lifespan of treated cells is measured in hours. In this
proposal we describe an alternative method which has the potential to overcome
many of the problems described above. The method involve; connecting mammalian
nerve cells to integrated electronic circuitry using extracellular
electrodes. Although in principle this is not a new idea, use of technique to
date has required a high level of experimental expertise and of specialized
silicon processing. The non-invasive nature of extracellular
electrodes allows long-term measurements to be made.