Ultrasonic transducers used as sensors for any kind of inspection invariably require signal processing of the output signal. The same is true for virtually all simple sensors. The minimum processing is usually amplification and filtering but often additional signal processing is needed. If you have sensors of this kind and you want to improve them for a negligible or a relatively small increase in component cost by adding intelligence then Cambridge Ultrasonics would like to help you. If you are not planning to add intelligence to your sensors then watch-out, your competitors may beat you to it.
A new opportunity has developed and it is all thanks to the large market for mobile telephones. The requirement for low-cost mobile phones has resulted in inexpensive but high performance digital signal processing (DSP) devices becoming available. Inside every mobile telephone is a DSP and peripheral devices that cost about $ 3 in volumes of about 1,000 - this is roughly the same as the cost of the linear components found in the analogue stages behind most sensors. So it is now possible to increase very substantially the degree of interpretation behind a sensor using digital signal processing and transform it into an intelligent sensor at a cost commensurate with the cost of existing components.
What kind of new functionality and interpretation is available? In one sense the sky is the limit but let’s consider more mundane functionality first. DSPs are designed with digital filtering in mind so low pass filtering, high pass, band-pass, notch-reject, lock-in amplifiers (notch-accept) and high-order filters are all possible - these can all be implemented as analogue circuits but DSPs can do the job more effectively and often more cheaply. For example, implementing a stable, sharp notch filter, say with a narrow-band centred on 50 Hz, using analogue circuits requires matching pairs of high tolerance capacitors and resistors and careful adjustment if the centre frequency is not to wander. Getting better than 2% (1 Hz) stability is difficult and it quickly becomes expensive to do better. But with a DSP the circuit design is straightforward in principle, there is no tuning or adjustment necessary, no special components are needed and the system remains as stable as the quartz clock controlling the ADC, typically at 2 ppm per day and 2 ppm per degree Centigrade - so the centre frequency should not wander by more than 0.1 mHz.
However, DSPs can do much more than just filtering. Adaptive filtering is possible, fuzzy logic can be implemented, artificial neural networks for decision making, communications and user interfaces are all possible. It is even possible to make a sensor that communicates over a network, for example Ethernet running TCP/IP so that the sensor looks like a web-page. The DSP can be programmed to handle the TCP/IP protocol for collision avoidance. In this way you can have a network of sensors, they don’t have to be the same type of sensor, and provided each has its own address one can collect information using standard networking procedures - it’s even possible to use Internet Explorer to view the page for each sensor.
More and more sensor technology is being developed in this way. A good example is CAN-bus which was developed to network many of the sensors and controls in automotive vehicles and thereby simplify the wiring and save weight in the vehicle. In this case every sensor and actuator requires a degree of intelligence.
At Cambridge Ultrasonics we are working on a project of this kind, using a low-cost DSP made by Texas Instruments to provide intelligence and communications so that the task of talking to sensors becomes simpler. We have always used DSPs in the past in high value systems due to the relatively high component cost, for example our first DSP project in 1993 was an ISA plug-in board for a PC to make a novel lock-in amplifier (we are still using one of those ISA boards ourselves to measure the vector impedance of piezoceramic components). Ten years later the new DSP has 1000 x the performance of the old DSP but is only 1% of the cost!
Another inherent advantage of having simple analogue processing with low-cost DSP processing is that changes to the product become software changes only. If the sensor is networked it may even be possible to up-grade the software remotely while the sensor is still in service.
The main disadvantage of using DSPs is the development cost, you need: hardware development systems, compilers for the different DSPs, logic analysers, high performance oscilloscopes and experienced staff. We also use Matlab for prototyping of signal processing algorithms. This is a significant extra cost but it saves time on each job we perform.
If you want to add intelligence to your sensors at low component cost then Cambridge Ultrasonics is available to help you.
Cambridge Ultrasonics is a source of ideas and innovation for its clients. Using our knowledge and experience in ultrasonics, physics, mathematics, electronics and software we work with our clients’ R&D departments to help them find solutions to their problems. Our flexible approach to working results in our clients effectively filling gaps in the skills profile. Over the last 15 years we have worked with large blue chip companies through to small emerging businesses, government agencies, research institutes and universities. Most of our clients return to make use of our services over the years.
We help our clients in the following ways:
The University of Manchester Institute of Science & Technology (UMIST) is looking for a commercial partner to help it in developing and exploiting a novel method for detecting the early stages of cancer of the mouth. The partner would ideally be an existing manufacturer of medical ultrasonic equipment that could exploit the system commercially.
Cambridge Ultrasonics has been appointed as a consultant on the project by UMIST. Our job is not only to advise UMIST on the ultrasonic and systems design but also to provide some commercial assistance. So this is an example of a client for whom we are providing a dual service of technical advice and commercial mentoring. This article in Innovation News is one way of helping.
UMIST wants to develop an ultrasound probe specifically for use inside the mouth. The main area of application is the imaging of oral cancer and pre-malignant conditions. This requires a high frequency, wide bandwidth, ultrasound probe operating with a centre frequency in excess of 20 MHz. A new ultrasound phantom (test target) will also be developed since existing phantoms are designed to test probes operating in the range of 3 to 12 MHz.
Diagnostic ultrasound systems which are commercially available at present do not operate above about 13 MHz. Commercial probes are, in any case, far too large to be used comfortably inside the mouth. Although no dedicated intra-oral probe exists, there is evidence that 20 MHz ultrasound would be best in this applications
In the UK, there are nearly two thousand new cases of oral cancer each year. Oral cancers represent about 1% of all malignant disease but the incidence is rising. In other countries, the prevalence is even higher, for example India, where oral malignancy constitutes 40% of all cancers. Most oral cancers (>80%) are squamous cell carcinomas. In the UK, about 900 deaths occur each year. This is a rate similar to that from breast cancer and cervical cancer. The five year survival rate is poor, at 30% to 50%, and the survival rate is further reduced if cancer spreads to lymph nodes in the neck. As in all cancers early detection is critical and that means detecting small cancerous sites.
UMIST plans to create small probes, about the size of a toothbrush, to scan the tongue and other parts of the mouth to find the all-important small cancerous sites. Dr John Hatfield of UMIST is in charge of the project, his specialization is in fabricating unusual micro-electronic devices. The plan is for the toothbrush-sized probe to have many of the electronic circuits built into it. At first sight this might seem unnecessary but there are good technical reasons for adopting this approach, for example, the capacitance of a receiver sensor in the probe will be small because of its small size so having its pre-amplifier nearby helps reduce cable capacitance that would otherwise severely attenuate the signal, likewise having the drive circuit for the transmitter close to the transmitter helps to reduce electromagnetic-interference and cross-talk in the receiver circuit. These advantages should all contribute to better quality images and improved detection of cancer sites.
The new colour printing of Innovation News allows us to show some of the posters we have designed and produced for the Institute of Physics over the last few years. Two of the events have been held just a few days before Christmas with the objective of making science more fun. David Andrews of Cambridge Ultrasonics has been heavily involved with these events.
This Christmas David is resting from IoP duty but he is still trying to promote physics by becoming a Science Ambassador in two Cambridge schools. Science Ambassadors are experts who visit schools to give lectures, help with projects and science clubs, give advice on equipment, careers and generally show that there is an alternative to life as a pop-start or professional footballer. For more information on Science Ambassadors see www.setnet.org.uk
 |
 |
 |
 |
Pedro Yip and Owen Budd joined Cambridge Ultrasonics this year. Owen used to work for a large consulting business in Cambridge; his experience is in hardware and software design. He is currently working on a circuit with a low-cost DSP, usually to be found in mobile telephones, but to be used in his application as a novel, intelligent ultrasonic sensor (see page 1).
Pedro is studying Engineering at Cambridge University. He worked over the summer with Cambridge Ultrasonics to get work experience and to learn more about electronics. He has been working on a project involving programmable logic devices and writing DSP software - Owen was giving some assistance. Pedro returned to start his second year of studies in October. We shall miss his enthusiastic personality and everyone at Cambridge Ultrasonics wishes him well in his studies and future career.
Cambridge Ultrasonics is looking to recruit a scientist/engineer preferably with a PhD in physics or engineering with particular strengths in electronics and software engineering. Software skills should include Matlab and C++. A good understanding of hardware would be an advantage. Experience of ultrasonics or sonar or radar is also desirable. More details on www.cambridge-en.com.
An acoustic or ultrasonic wave can lose energy by two principal mechanisms: thermoelastic losses in homogeneous materials (mechanical energy is converted into heat) or scattering in heterogeneous materials (where the wave loses coherence and is scattered in many directions). Both mechanisms result in an apparent loss of intensity of the wave, which is commonly referred to as attenuation. But things aren’t as simple as they seems because scattering does not convert ultrasonic wave energy into another form of energy so little or no energy is lost.
Theory shows that the attenuation value is proportional to frequency squared for thermoelastic losses but for scattering losses experiments show it is directly proportional to frequency or some non-integer power.
The way attenuation is measured can strongly influence the result when scattering is the dominant mechanism, which implies the attenuation value is not a true material property. Doubtless researchers working in standards laboratories accept this situation as an everyday occurrence.
Concrete is a good example of a scattering material - it’s a material we have worked on extensivley at Cambridge Ultrasonics. The randomly distributed stone or aggregate particles in concrete make it a random scatterer if the wavelength is of the same order or less than the aggregate size. If attenuation is measured using a matched pair of large aperture transducers (large in the sense that the aperture is much larger than a wavelength) then the transmitter will launch plane waves and the receiver will register a small signal, indicating high attenuation in concrete. Performing the same experiment on a homogeneous material, like aluminium, results in a much larger received signal. Concrete appears to have a larger effective attenuation than aluminium for this particular experiment but it is not correct to conclude that the attenuation value measured is a material parameter that can be widely applied to other experiments, as we will now discover. If an energy detector, made from an array of small aperture receivers, is used instead of the single, large aperture receiver then the attenuation of concrete is found to be approximately the same as the attenuation in aluminium.
An energy detector adds together the (unipolar) energy values of the outputs from the array rather than the (bipolar) voltage signals. Energy is always positive in value (unipolar) so one way to make a simple energy detector is to full-wave rectify the output of each receiver, smooth the signal then sum over all the elements of the array. This eliminates the possibility of destructive interference and allows only constructive interference when summing. The effect of random scattering is now substantially reduced, merely by changing the way that the intensity is estimated.
Clearly, it is unsatisfactory that the attenuation value depends strongly on the experiment and illustrates that attenuation by scattering must be considered with careful reference to the experiment.
Each scatterer in a heterogeneous material can create three waves during scattering: a reflected wave, a diffracted wave and a transmitted wave. All three components introduce time delays, or phase differences, that depend upon the size and material properties of the aggregate and since size and material properties are to some degree randomly distributed this is how wave-fronts become progressively, randomly disrupted as they travel.
A distribution of wavelets is created with a distribution of random phases or time-delays relative to the unscattered waves because waves are scattered by the randomly arranged aggregates. Scattering has an effect that is similar to dispersion. The randomness can be caused either by multiple scattering, so that the path length is increased, or by wave speed differences between the aggregates and the cementitious matrix. The result is that the initial, narrow distribution of phases in the coherent wave-pulse become a wider distribution of randomly phased wavelets arriving at the receiver. A large aperture receiver with a single output integrates all the contributions over its surface. The distribution of phases over the aperture results in destructive interference almost as often as constructive interference and the greater the degree of scattering the more closely the electrical output approaches zero. However, the small aperture receivers in the array can sample individual wavelets without summing to zero if the aperture is no greater than one wavelength.
In medical diagnostic applications random scattering is referred to as speckle (a term first coined in laser physics). Speckle appears as noise in the compound B-mode images commonly used for medical diagnosis. There are techniques used for reducing the speckle in images: frequency compounding, spatial compounding and more conventional image enhancement methods. It is also possible to use statistical methods to apply a time-dependent threshold level to image signals and only allow signals greater than the threshold to contribute to the brightness. The Rayleigh distribution can be used but the Weibull distribution is generally found to give better results. Similar methods are used in radar systems.
Attenuation is formally measured in nepers
m-1 but it is more commonly measured in dB m-1 or sometimes in
dB m-1 Hz-1. The neper is a consequence of the exponential decay
of intensity with distance.
(db m-1) = 20(log10e) (neper m-1)
It is also common to compare materials in terms of attenuation per wavelength (measured in dB) because the size of an experiment or ultrasonic system is usually an important underlying factor to consider and, generally, the size of the experiment scales with the wavelength.
All materials become grainy at sufficiently small scales: concrete is grainy on a scale of the order of 10 mm, metals and alloys (putting aside amorphous metals) crystallize to form interlocking grains with sizes between a few microns to about 1 mm, water has a graininess on a molecular scale but at larger sizes there are impurities and micro-bubbles of gas present.
Once the wavelength is commensurate in size with the natural graininess of the material then scattering becomes an important effect. When scattering starts to be important then the size of the receiver will strongly affect the signal you want to process and the apparent attenuation of the material. Don’t expect the text-book value for attenuation to be accurate in predicting the attenuation in your experiment or equipment - the author of the text-book doesn’t know what size of receiver you are using.
Figure 1 - Sketch to illustrate the effect of scattering losses
in an attenuation measurement experiment and the effect of receiver
type on the result. The homogeneous material produces a measurable
signal with a simple receiver but a heterogeneous material gives
a low output (A) with a simple receiver due to apparent attenuation.
However, the heterogenous material can still give a greater signal
(B) with an energy detector. The signal processor used with the
array receiver is either a summing circuit allowing destructive
interference (output A) or a summing circuit that does not allow
destructive interference (output B).
Figure 2 - A time-frequency representation of a linear sweep frequency chirp that has travelled though concrete. Colour is used to represent amplitude. The yellow region shows the chirp, which appears as an bar at an angle to the time axis - it’s slope is the rate of change of frequency of the chirp. Note that energy has been scattered to later times. The slope of the chirp changes and this is one of the effects of dispersion.
Figure 3 - Photograph of waves rendered visible using Cambridge Ultrasonics’ schlieren visualization equipment. Waves in water are scattered by a single, solid cylinder.
Figure 4 - Photograph of waves visualized in a model of concrete. The effect seen in the figure with a single cylinder is repeated with several cylinders. The random position of the scattering cylinders introduces randomness into the wave pulse but no loss of energy.