Sensors

The sensors used in this project had the task of collecting information relating to ambulation of a subject, which was then recorded and transmitted to the software for processing.

A number of different sensors and configurations were trialled, predominately concentrating on collecting information from the lower leg and foot. Two primary types of sensors were used; accelerometers and pressure transducers.

Compumedics supplied some of their manufactured piezoelectric accelerometers, designed to be used with the Siesta to detect movement during sleep recordings. As such, these sensors were designed to be strong, cheap and produce qualitative information relating to movement.

After examining the piezoelectric accelerometer response, it was decided that some different sensors for producing quantitative information relating to foot up or down state - pressure transducers - were required. This chapter discusses the results of investigations into the piezoelectric sensors' response, their applicability to ambulation monitoring, implementation of the pressure transducers and results from other sensors which were briefly tested but not used.

   
Piezoelectric accelerometers

The piezoelectric accelerometers supplied by Compumedics utilise the piezoelectric properties of a polarised polymer. When the polymer is disfigured, it produces a small voltage which is recorded by the Siesta.

Piezoelectric effect

The term `piezo' derives from the Greek `pressure' or `to press' and is defined as ``polarisation in a substance resulting from application of mechanical stress, especially in certain crystals'' [56]. The piezoelectric effect occurs in crystals such as quartz, which produce small voltages across its faces when subjected to mechanical pressure [13]. Conversely the piezoelectric effect works in reverse - applying a voltage to a piezoelectric substance causes the crystal structure to deform, resulting in movement.

Due to these effects, piezoelectric devices find wide use in electronics applications such as high-frequency oscillators[57] and transducers. Applications such as accelerometers, microphones and audio speakers usually use a specially treated polymer film which produces voltage between two points (or movement) when subjected to pressure (or a voltage).

  

Figure 6.1: Compumedics supplied piezoelectric sensor and connecting leads

Placement and sensor response

The response of the piezoelectric accelerometers was predominately AC (high frequency), with little to no DC component. An example of the accelerometer response is shown in Figure 6.2. This data was recorded from sensors attached to the top of each shoe while taking two slow steps, and shows peaks when the foot starts to swing and then strikes. It was discovered that in practice it was very difficult to reliably distinguish between a foot-strike and toe-off (Section 2.1.1), or extract distinct events from the multiple signal peaks usually resulting from a single movement. These sensors did provide useful evidence of fast and sharp movements however, such as a hard footfall that may occur when correctional steps are utilised to maintain balance.

  

Figure 6.2: Example piezoelectric accelerometer response

Another method of using the piezoelectrics trialled was to place them under the in-sole in the shoe to measure pressure as weight is applied and removed from each foot (see Figure 6.3). The sensor response tended to give large `spikes' when the load was applied and removed. Again, this produced problems with distinguishing between foot up and down.

  

Figure 6.3: Response of piezoelectric sensor placed in shoe in-sole

Applicability

After much work with the piezoelectrics and attempts to extract concrete ambulation information from the signals, it was concluded that the piezoelectric sensors alone could not provide enough quantitative (and to an extent, even qualitative) information for the project's purposes. Although not particularly well suited for providing evidence relating to the subject's stage in the gait cycle, or whether the feet were on or off the ground, they were still considered useful for detecting the accelerations that may occur during a stumble of fall. The high-frequency noise and interference of the sensors could be handled using a combination of software filtering and an `unreliable' sensor model in the DBN.

FSR pressure transducers

In the previous project by McGowan [4], force-sensitive resistors (FSRs) were successfully used as pressure transducers, but it was discovered similar sensors are very difficult to come by in Australia. Some investigation lead to Associate Professor Andy Russell from the Department of Electrical and Computer Systems Engineering, who through his work with tactile sensors for robotics possesses a wide array of pressure-sensitive resistors and materials. He provided samples of some FSRs and a sample of material called `Vermahide'. The delicate nature of the manufactured FSRs resulted in the design and implementation of some new pressure transducers using Vermahide.

McGowan noted in his thesis that the pre-packaged force-sensitive resistors used in his project were liable to breakdown as a result of the large forces exerted on them when used as foot sensors. Due to the increased likelihood of failure in these sensors, compared to the relative robustness, flexibility and adaptability of the Vermahide material, it was decided that pressure transducers involving Vermahide would be more suitable as foot-sensors. As a result a variety of sensors involving different configurations of Vermahide and electrical contacts were built and tested.

   
Vermahide pressure transducers

Vermahide [58] is a felt-like fabric composed of openly-matted synthetic fibres, which are encased in a conductive carbon emulsion. The emulsion itself is not pressure-sensitive, but when pressure is applied to the material, the fibres move closer together, and their effective conductive paths from one side of the material to the other are reduced, resulting in a drop in resistance. The matted fibre structure of the material gives it quite a good elastic memory, in that it will tend to restore to its original shape after pressure is removed.

Preliminary investigation of Vermahide pressure transducer

Some initial sensors developed using aluminium contacts sandwiching a piece of Vermahide; this worked well but it was found that the leads to the sensors did not solder well to the Aluminium, and could easily be pulled off. Under the advice of Professor Andy Russell, the use a of printed circuit board (PCB) was trialled. The first Vermahide-PCB sensor used a pre-printed PCB with 2 mm wide parallel copper tracks, separated by 1 mm. The two connecting leads for the sensor were electrically connected to alternating copper tracks by way of short soldered connecting leads. The Vermahide was sandwiched in between the PCB and a stiff non-conducting material (cardboard was used) so that it contacted the tracks. Hence if pressure was applied the resistance measurement was only taken from one side of the Vermahide.

The alternating track connections required delicate wiring and soldering, increasing the risk of the sensor becoming a short circuit, resulting in large signals which may damage the Siesta.

Design of final pressure transducer

In an attempt to overcome the above shortcomings, the final design use a $20\times20$ mm of Vermahide between two pieces of one-sided plain copper-plated PCB. A small hole was drilled through the PCB so the lead could be passed through from the un-plated the to plated side and soldered to the copper which was in contact with the Vermahide. As in the aluminium-contact sensors, the Vermahide was sandwiched between two pieces of PCB so that the resistance across the material could be measured, as shown in Figure 6.4. The whole assembly and protruding leads were secured with electrical tape to keep the sensor's thickness to a minimum. This configuration was used for building more sensors, as it appeared the most robust and simple implementation, reducing the risk of sensor failure due to short circuits or disconnected leads. During testing, no such problems were encountered with these sensors.

  

Figure 6.4: Exploded view of Vermahide pressure sensor

   
Circuit design

As the Vermahide sensor changes resistance with pressure, but the Siesta expects voltage inputs, a small circuit was required to produce a voltage relative to the change in the sensor's resistance due to applied pressure. This circuit was a voltage divider, consisting of power source V_S (a 1.5 volt AA battery), a fixed resistor R and the foot sensor R_F, which acts like a variable resistor.

  

Figure 6.5: Voltage divider circuit diagram, where `Rf' represents the Vermahide sensors and `R1' to `R4' are the fixed biasing resistors.

As each of the four Vermahide-based sensors needed to give independent signals, the circuit based on the diagram in Figure 6.5, consisting of four parallel voltage dividers (one for each sensor), was implemented. The four fixed resistors, sensor leads and connectors for the Siesta datalogger were soldered to a piece of pre-printed circuit board and placed in a small containing box with the battery to prevent damage.

  

Figure 6.6: Four Vermahide pressure sensors, box containing voltage dividers, `AA' battery, and connecting leads. Siesta shown with back cover removed, exposing batteries and input channel connectors.

Given the voltage source value V_S, the resistance values of R and the sensor R_F, the output voltage to the datalogger V_O can be calculated using the voltage-divider formula (Equation 6.1). This equation assumes that the input impedance to the Siesta datalogger is so high that it may be considered an open circuit, especially in relation to the output resistance of the circuit in Figure 6.5. The sensor didn't need to accurately detect the amount of pressure being applied, merely the relative difference in pressure for different situations, so some deviation was acceptable.

 

$$V_O = \frac{V_s \times R_1}{R_1+R_F}$$ (6.1)

$$\therefore R_1 = \frac{V_O \times R_F}{V_S - V_O}$$

The fixed resistance values for R in each of the voltage dividers were selected by re-arranging Equation 6.1, with V_O set to the maximum required input signal voltage to the datalogger, and R_F the smallest expected sensor resistance. To reduce the risk of damage to the datalogger, V_O was chosen to fall within the datalogger's second largest gain of 100 millivolts peak-to-peak^6.1.

The resistance ranges of the $2\times{}2$ cm Vermahide sensors built were measured so appropriate voltage divider R values could be calculated. Typical value ranges are shown in Table 6.1. The voltage divider circuit produces the largest signal voltage for the smallest pressure sensor resistance, and the minimum value of R_F (including a safety margin) under maximum load was taken to be around 15 to 20$\Omega$.

Using these resistance readings, the voltage divider formula and practical trials (results shown in Figure 6.7), it was established that suitable voltage divider biasing resistances were in the range $680\Omega$ to $1k\Omega$. As different sensor placements were subject to different amounts of pressure, and larger R values produce larger signal voltages for less pressure, it was decided to implement two voltage dividers with $R=680\Omega$ and two using $R=1k\Omega$. The $680\Omega$ voltage dividers were used with sensors receiving the most pressure, as is found under the heel, and the $1k\Omega$ voltage dividers were used with sensors placed under the toe. This resulted in all the signals varying over a similar range in response to their range of applied pressure, which in turn helped produce simpler and more reliable feature extraction.

 

Table 6.1: Vermahide heel pressure sensor resistance measurements

Applied Load	Measured resistance
No load (in air)	1M	-	1.5M$\Omega$
Placed in shoe in-sole	100k	-	200k$\Omega$
Body weight (double support)	30k	-	40k$\Omega$
Body weight (single support on sensor)	20k	-	30k$\Omega$

 

Placement

It was decided that the best pressure sensor positions for collecting useful ambulation information was under the heel and toe, as these positions play crucial roles in the gait cycle. The heel is (usually) the point of contact for a foot-strike, and the toe-off event marks the start of the single-support phase (See section 2.1.1).

Heel sensors were placed as close the the centre of the Calcaneum (heel bone) as possible, so that it would receive most pressure during mid stance (tibia vertical) phase. Two toe sensor positions were investigated; under the base of the toes (between Metatarsals and Phlanges joints) and under the big toe.

Sensors were either placed under, or attached to, the shoe in-sole using tape. The in-sole sits between the foot itself and the sole^6.2 of the shoe. Although the sensors were not in direct contact with the foot, the foam in-sole tended to `mold' around the sensor, discouraging sensors from moving laterally and out of position. Additionally, this method was found to be more comfortable for the subject.

During trials, it was noticed that the sensor response relied heavily on placement - if a heel sensor was situated towards the side of the foot, it would produce smaller signals, which may result in significant gait cycle events being missed. Although DBNs can compensate for observation errors through sensor models and conclusions made from other evidence, it was still preferable to keep sensor readings as accurate as possible.

Response

The sensors and accompanying circuits were designed to produce a peak voltage around 50 mV for maximum pressure (to suit the datalogger's 100 mV peak-peak gain setting), and less than 10 mV for no pressure. Table 6.2 shows some typical voltage readings taken from the sensors under different loads.

 

Table 6.2: Vermahide sensor and circuit voltage response

Applied Load	Output voltage
No load (in air)	0.2m	-	3 mV
Placed in shoe in-sole	3m	-	6 mV
Body weight (double support)	15m	-	30 mV
Body weight (single support)	30m	-	50 mV

 

Some factors influencing sensor response included placement, properties of the shoe and in-sole, battery voltages, body mass and physiological parameters of the subject. These factors and the inherent physical differences between individual sensors meant the signals were not identical. Accordingly, after sensors were placed, their responses were tested by standing in double support, then concentrating pressure over each of the sensors in turn. The results for such a test is shown in Figure 6.7, where he difference in signals for no sensor load, double support and concentrated pressure over a sensor can be seen clearly. As long as there were distinct differences between low and high pressure, the software could be calibrated to trigger `up' and `down' evidence for the appropriate signal voltages.

  

Figure 6.7: Annotated Vermahide pressure transducer signal test

Electromyograph (EMG)

Some trials were performed with Jean McInerney (see Section 1.4.1), to test the possibility of using electromyograph (EMG) signals for collecting ambulation-related muscles activity information. EMG sensors pick up the electrical activity which occurs within muscles when they move via electrodes coated in special conductive gel placed on the skin above the muscle of interest. Trials were performed with sensors attached to calves and and thighs. Initial test results are shown in Figure 6.9.

  

Figure 6.8: Compumedics piezoelectric (left) and EMG sensors (right)

  

Figure 6.9: EMG trial signals (top), piezoelectric accelerometer signals from foot (bottom)

After some trials, it was decided it was not worthwhile continuing with EMG sensor trials for a number of reasons. Firstly the EMG sensors are quite difficult to set up, requiring ground and reference electrodes - usually attached to the neck or head - as well as those attached over the muscles. Reference electrodes have to be placed carefully so they can be used to remove unwanted signals and interference from elsewhere in the body. Secondly, the EMG signals are characteristically very small in magnitude - in the range of 500 microvolts, and are susceptible to interference from other organs, external power supplies, crosstalk or capacitance between electrode leads and lead or sensor movement.