2. METHODS 
2.1. Measurement System using two Accelerometers 
This system consists of two ADXL (ANALOG DEVICES) 
accelerometers, which are dual axes respectively that are 
connected easily to a small WPC of a wristwatch type (Figure2). 
The WPC controls the synchronization of two accelerometers, 
which have a ceramic resonator, called a *SCERALOCK", 
respectively. Lithium rechargeable batteries supply electrical 
power to the two sensors. While walking, the WPC is worn on 
a subject's wrist, the sensors are fixed to a subject's left and 
right knees, and batteries are inserted into a subject's waist 
pouch. When a switch is turned on, the serial time counts and 
accelerations in two axes of subject's both knees are recorded 
into the WPC at 15 Hz. 
measurement is 12 sec +5 %. A person can walk everywhere 
The accuracy of acceleration 
naturally, because long cords are not necessary between an AC 
power supply and computer and the person. 
Waist Pouch 
Batteries 
Accelerometers 
  
Figure 2. Measurement System using Accelerometers 
2.2. Measurement System using Gyro Sensor 
This system consists of a MAXCUBE gyro sensor and a WPC 
(Figure 2). The output of roll and pitch angle is limited in + 
180 ^ , pitch angle is limited in + 90 ° The accuracy of 
measurement of each angle is under 0.2 ^ . When a switch is 
turned on, the calibration of MAXCUBE gyro sensor starts 
before measurement. After 10 minutes, an end sign appears, the 
measurement starts by pushing a button of WPC. The degrees 
of angles become 0° just after calibration and the serial time 
count, their relative data are acquired. When a switch is turned 
on, the serial time count, relative three angles, i.e., roll, pitch 
and yaw meterages of the patient’s back are recorded into WPC 
at 30 Hz. They are also displayed on a liquid crystal display on 
the WPC. 
2.3 Measurement System using Pulse Sensor 
This system (Pulse Graph, SEIKO) is a lightweight, portable 
and useful for a training tool of walking or running. It consists 
of LED and WPC (Figure 3 ). The supporter of LED covered a 
patient's forefinger. It is possible to measure the real-time pulse 
rate during movement. The blood flow was sensed by LED and 
was converted into the pulse rate at intervals of 4 seconds. 
  
Figure 3. Measurement System of Pulse Sensor 
2.4 Spectral Analysis using AR modeling 
Multivariate AR modeling is given by (1), 
x;(s)=} > a,(m)x, (s-m)+u,(s)  q) 
J=1 m=1 
where X;(S) = stationary time series 
X;(s-m) = past observed data 
u;(s) = white noise 
aj(m) — AR coefficient 
The frequency response function aj(f) of x;(s) to the input x;(s) 
is given by (2). 
M 
a,(f)=} a,(m)e 7 Q) 
m=l 
where — e "7?" = Fourier transform of frequency response 
The system given by (1) is a feedback system within which X;(s) 
is connected to x;(s) by an element having the frequency 
response function a;(f) and each x;(s) has its own noise source 
u;(s)'s. Thus, x;(s)can be expressed as a sum of the influences of 
u;(s)'s. The estimate of the power spectral density p (f) is given 
by (3), 
p OM) 
  
zs 2 (3) 
1— > a(m)e "7?" 
m=] 
where  6°(M)= covariance 
This formula defines the best prediction, giving the power 
spectral density, which expresses the characteristics of a 
sequential system concisely, decomposing it into periodic 
components. 
—282—