Criteria for Setting Switching Levels in Myoelectric Prostheses
R. N. SCOTT, P A. PARKER, P A. O'NEILL ANDE. L. MORIN
In a myoelectric control system the signal from surface electrodes is processed by electronic circuitry, after which an estimate is made of the signal level. That estimate of signal level is used either to select a function or to control the speed of a function.' In the conventional Otto Bock system (with MyoBock single channel electrodes), and in myoelectric prosthesis systems made by Hugh Steeper Ltd., by the University of New Brunswick (UNB) and by Variety Ability Systems Inc., the decision as to whether a function is to be activated and/or which function to activate depends upon the signal level. In all cases, such systems provide some means to adjust the signal levels) at which these decisions are made.
The processed myoelectric signal corresponding to a constant contraction is not itself constant. Rather, the processed signal fluctuates above and below a mean value corresponding to that contraction level. In order to obtain an estimate of the contraction level, it is necessary to observe the signal for some period of time. The more time allowed for this process of estimation, the more precise will be the estimate. But, for the system to be of clinical value in controlling a prosthesis, the estimation process must not take much more than one-tenth of a second. Otherwise, an unacceptable delay will occur between the amputee's muscle contraction and the resulting prosthetic function. Consequently, considerable uncertainty exists in the resulting estimate. In a typical system, for a moderate contraction, the estimate of myoelectric signal level will be within about 20 percent of the true value. Accordingly, there is a chance that the system will not act in the way that the amputee intended, even though the wearer did produce exactly the intended signal level. The probability of this happening is called "system error".
The task facing the amputee in using a myoelectric prosthesis is difficult. It involves generating a myoelectric signal level corresponding to a specific target (or within a specific range) without immediate feedback as to how the signal corresponds to that target. There is feedback, of course, but only after the prosthesis has been activated-possibly in the wrong direction. Normally, the amputee produces a signal which is somewhat different from the target. The difference between the signal actually generated and what was intended is termed "operator error". Thus the actual behavior of the myoelectric prosthesis results from the intent of the amputee modified by two sources of error -system error and operator error. Clearly, these errors are not prohibitive in the simple systems which have achieved clinical acceptance. They are responsible for the decision of the UNB group not to implement a five-state system, which would, in theory, provide on-off control of prehension and wrist pronation/supination from a single muscle site. But speed of learning, ease of use, and possibly accuracy of use, depend upon setting the decision levels in the control system so as to minimize the problem.
Neither system error nor operator error can be eliminated. What can be done, and should be done, is to select targets for the myoelectric signal which minimize the chance of an incorrect response. This involves determining the range of signal levels which the amputee can generate comfortably (the dynamic range) and choosing optimum target(s) within that range2 Unfortunately, determining the dynamic range also involves an estimation process and associated errors. Because estimation is done by clinical personnel, unlike the estimation performed electronically in the control system, it is important to explain the estimation process in order to minimize errors at this stage of the process. The procedure requires estimating two signal levels, the minimum signal and the maximum signal.
Estimation of Minimum Signal
Estimation of the minimum signal is deceptively difficult. What is sought is not the signal measured at the electrodes with the amputee fully relaxed, but rather the signal which is likely to be observed when the amputee is engaged in a full range of normal activities but not contracting the muscle(s) used to control the prosthesis. The problems are obvious. How energetic should the amputee's activities be during testing? Should one record the highest signal observed, even if it lasted only an instant, or should one attempt to judge some sort of average level of signal? If a prosthesis is not being worn when this measurement is made, how may it be expected to alter the signal levels? Some environments are electrically more noisy than others: should a particularly noisy environment be used for this measurement?
The prosthetist or therapist should measure the minimum signal under conditions which are as close as possible to those in which the prosthesis is to be used. After all, the purpose is to determine the background signal which is most likely to exist under normal circumstances when the amputee is not attempting to activate the prosthesis. Extremely brief peaks of signal during physical activity may be ignored, as they are unlikely to activate any prosthesis. The highest signal consistently observed, rather than the average, should be recorded. And if the amputee will be using the prosthesis in a location known to generate high levels of electrical interference-near an electrical substation or radio transmitter, for example-measurements should be made in that environment, if possible. If the estimate of minimum signal is substantially lower than the actual value, the amputee will have trouble preventing involuntary activation of the prosthesis. In the extreme case; it may be possible to prevent activation. If the estimate is substantially higher than the actual value, the effect is to use only a portion of the dynamic range of the signal and thus to require more precise control by the amputee for all functions.
Estimation of Maximum Signal
Estimation of the maximum signal level also is challenging. What is sought is the signal level which the amputee can sustain comfortably during the time required to activate the prosthesis, throughout the day, without significant fatigue. Clearly, it is not feasible to spend a day measuring this, so the clinician must make an estimate based upon how the amputee performs over a relatively short time. Most prosthetic actuators controlled by on-off myoelectric systems require only one or two seconds of continuous activation, but the frequency of use will affect the amputee's ability to sustain a high level of contraction without fatigue. We recommend that the amputee be asked to produce a strong signal which the wearer finds not uncomfortable over a period of about 5 seconds. Normally the signal level at the end of this time is a reasonable estimate of the signal which the amputee can achieve throughout the day without tiring.
If the estimate of maximum signal is too high, the amputee will experience difficulty activating the prosthesis later in the day even though performance seems satisfactory initially. If the estimate is substantially lower than the actual value, the effect is to use only a portion of the dynamic range of the signal and thus to require more precise control by the amputee for all functions. When the UNB Universal Myoelectric Trainer is used to estimate minimum and maximum signal levels, the user has the option of allowing the Vainer to average the signal over a few seconds. This is helpful in avoiding very short duration peaks, but it does not relieve one of the necessity of ensuring that the signal observed by the Trainer really is a representative sample.
Once the maximum and minimum signal levels have been established, the decision threshold(s) for the control system should be set at optimum values. These have been established on the basis of analysis of the probability of incorrect prosthesis function due to system and operator error. Using the proper settings will minimize the chance of improper function and, most significantly, make the prosthesis appear to be easier to control. Because the algorithm for selecting decision thresholds is quite complex, tables have been prepared to guide the user.3 Alternatively, the UNB Universal Myoelectric Trainer may be used, in which case the settings are computed automatically.
In a system which has only two conditions, either "on" or "off", (what is referred to in UNB product information as a two-state system) a single setting is required to determine how much signal is required to activate the prosthesis. This may be called a "gain control", which is equivalent to a radio volume control and affects the amount by which the signal is amplified. It may also be called a "switching level control", which actually changes the signal level at which the prosthesis will be activated. In a system where a small signal activates one function and a larger signal activates a different function (what is referred to in UNB product information as a three-state system), two settings are required. Usually these are actual switching levels.
Clearly, the adjustment becomes more critical as the number of active states is increased. In a two-state system, such as a one muscle "cookie crusher" arrangement or half of a two muscle on-off control (e.g. a standard Otto Bock setup) the amputee can tolerate an adjustment which is far from optimum without any noticeable effect upon performance. But even in that situation, the difficulty of the task for the amputee can be reduced considerably by making the adjustment more nearly optimum. Of course, for amputees who have a large dynamic range of myoelectric signal-a ratio, maximum to minimum signal, of at least ten-adjustment is not critical even for a threestate system. But myoelectric prostheses are used by many amputees who have much smaller dynamic ranges, including those who may have only a few active motor units in the controlling muscle. In such cases, the best possible adjustment is important even for two-state systems.
It is desirable that the amputee's task be made as easy as possible. To that end, decision thresholds (switching levels) in myoelectric prostheses should be set appropriately with respect to the available dynamic range of myoelectric signal. Experience has shown that this adjustment is complicated, in practice, by the difficulty of determining the dynamic range of the myoelectric signal. Estimates of both the minimum and maximum signal levels tend to be quite inaccurate unless the clinician has a sound understanding of the significance of these estimates and exercises good judgment. The procedure may be complicated by inadequacy of the equipment used to measure the myoelectric signal.
Institute of Biomedical Engineering, University of New Brunswick, Fredericton, New Brunswick E3B 5A3 Canada (Drs. Scott, Parker, and O'Neill). Department of Electrical Engineering, Queen's University, Kingston, Ontario, Canada (Dr. Morin).
Robert Caldwell, Sheila Hubbard, Karen Louison and Dinah Stocker provided clinical and experimental data which comprise the background for this paper. The authors' research in myoelectric control systems is supported by the Natural Sciences and Engineering Research Council of Canada and the Medical Research Council of Canada.
- Scott RN: An Introduction to Myoelectric Prostheses. UNB Monographs on Myoelectric Prostheses. A. Muzumdar, Ed., Vol. 1. Fredericton: Institute of Biomedical Engineering, University of New Brunswick, 1984.
- Parker PA, Stuller JA, Scott RN: "Signal Processing for the Multistate Myoelectric Channel," Proceedings IEEE 65:662-673, 1977.
- Tables of switching level settings provided with UNB MiniTrainer, available from Liberty Research, Hopkinton, MA and from the UNB Institute of Biomedical Engineering, Fredericton NB.