Processing the Electromyographic Signal


While seemingly technical, this short presentation illustrates the consistency with which Johnson applied his ideas: even his thoughts about prostheses are informed by the same respect for participation and letting the user set their own path as the rest of his writings.

I feel that the best service I can perform at this conference is to react to what I have seen here rather than to make a formal presentation which might be out of context. The primary impression I have had is that my own prior presumption of sophistication in the application of control theory to the man-machine interface has been considerably humbled. I’m impressed. I would, however, like to comment on servo theory as it might be applied to prosthetics, on some of the practical applications of this theory, and briefly describe the human himself as a signal processor operating within a system which has the capability of further, external signal processing.

When I was doing my Doctor of Philosophy thesis on the servo-analysis of human postural reflexes, I was appalled at the challenge of Imitating this servo system as an engineer put to the task of producing something like It. And if I hadn’t had the ideal model--namely, the human being himself—before me all the time, I would have said any number of times that it was impossible and given up. When one considers the compact power sources that we carry around with us and our system for control, both of fine movements and of ballistic ones, specification of an artificial substitute in terms other than physiological ones becomes incredibly difficult.

It is appropriate to keep in mind Sherrington’s observation that “all movement takes place on a background of posture.” One can, in fact, go further to state that all nonballlistic movements are pos­tural adjustments and are mediated by the neuromuscular feedback loops within and distal to the spinal cord. The observation suggests two things: first, the fairly obvious one that a prosthetic arm ought ideally to be capable of subsuming a role in postural support as well as a manipulative function; and second, in the performance of the latter role, the incorporation of closed peripheral feedback loops may be quite useful. If one is to avoid the problem regarding the requirement that the user of the prosthesis direct his attention always toward the output, perhaps the proper way to program a prosthesis is to program the end-point of the movement. That is, I can look at a pencil on the table behind me, look away, and then pick it up, getting my hand there by any one of a number of paths. What I program in my head is the position of the pencil, not a sequence of movements that must be carried out. This behavior requires feedback or “goal-setting” loops within the major loop of direct visual observation and the parameter to be controlled or programed is position. The possible exception is in the control of grasp wherein force or pressure on an object is the more important variable. In this case, position is automatically dictated by the geometry of the object.

The “ideal” prosthesis will require velocity feedback in a small inner loop so that one can limit maximum velocities and perhaps enhance the speed of some movements if the end-point is far away from the origin. But I would say that if you want to substitute in any sense for kinesthesis, for knowing where a limb is without looking at it, you have to program where it is going to be without looking at it. That is, the peripheral controlling signals must put out a statement of the end-point desired.

It is probable that in the practical application of signal-control theory to prosthetics one will not see any mass-produced inexpensive devices ready to wear. Industry, of course, wants to send something out the door which is, in some sense, foolproof and is general enough to have a wide market. It is apparent to me from what I have seen here that each prosthetics application must be specific to the individual to some extent. And this requires that the person applying the prostheses, one who is deciding what form it is to take, must ask each time, not “What does the patient want by way of output?” but “What will he want ten years from now” but “What does he want now?” As he acquires skill, as he ages, perhaps he may want to change the purposes of the prosthesis, but it is useless to reproduce one that is so general that nobody wants it.

Who decides upon this prosthesis? Generally the therapist who has daily contact with the patient and the technician who will build it. What is needed, then, more than an engineer constantly at hand, is a set of descriptions of available devices and materials in terms more physiological and anatomical so that the application may be patient-oriented as much as possible.

How much of this can come off-the-shelf from industry even if the prosthetics problems and their potential solutions are adequately described is still a fairly big question because we don’t know yet what an adequate description is.

The third and final area I would like to touch upon is that of the human himself as a signal processor. I think it is very important that no one ever expect that the use of a prosthesis be a skilled use the first time it is applied. I often get the feeling that when somebody is talking about a prosthesis he has applied he says almost apologetically: “Well, he was able to use it after practice; after weeks or years of practice.” I would like to ask any typist how long it took to learn touch typing or any athlete how long it took to acquire the skills that he or she has in using either the body itself or a prosthetic racket or club in the pursuit of that sport. A human, above all, is trainable.

Perhaps this suggests a next-step improvement in the system Case Institute has developed: make it trainable too. At present it stores a number of fixed patterns of movement from which the user may make a discrete choice. If the stored programs could be modified gradually—the parametric weighting functions being altered a little at a time through the approval or disapproval of the user—then in continued use the movement patterns would become more suited to the individual.

I think the future lies in making more controllable variables available on the user of the device and allowing him to modify those variables through his own learning: in other words, increasing the channel capacity at the interface. In this regard the work described by Dr. Marquardt, in which muscle bulges in discrete, unexpected places are elicited individually, points a way.

The human being is perhaps the best known mechanism for ignoring redundancies and for extracting signals from noise. It is incredible how well a human can perform as an information filter. As I work on gadgets for the blind and on prostheses, one request I always make is that the engineer please not try to recode information in a form that he thinks is simpler. Leave it in as raw a form as possible. The ultimate user can do the job of coding or interpreting far better and may, through practiced use, find information available that the inventor of the device never even suspected was there.

If one has a number of separate inputs to a prosthesis—from the subject to the mechanical system—I claim that orthogonality is not necessary. That is, the selection of signal inputs need not be independent; cross-talk is allowed. The human learns to get around this pretty well. The number of variables which are available to him need not be restricted to a simple set because, frankly, we don’t know what simplicity is. Simplicity may be a vast range of choice of parameters in which many are employed simultaneously in a weighted manner.

In conclusion, I would like to put forth two exhortations: one to the therapist—the person having direct contact with the user of the prosthesis, and one to the engineer who is to design the prosthesis or to build it.

To the therapist I would say: “Don’t be afraid to talk to the engineer. Describe the problem in as many ways as you can and, whenever possible, relate it to normal behavior as he can see it.” In that way the engineer can identify with the problem himself. He can try to imagine what it would be like to need the prosthesis himself. If the engineer does not now understand, maybe he is at fault and the best thing to do is to repeat what you have said already. I have too often encountered a description that was adequate but in terms unfamiliar to the listener—hence misunderstood.

To the engineer I would say: “Don’t be too elegant in the solutions that you offer.” It is no great sin to have cross-talk between channels and to have redundancy of function when it is not there for the purpose of reliability. The main qualities to provide are reliability and ruggedness, adjustability, range of function, and your own readiness to consider each problem afresh. We can’t sit back and say that we solved a problem well last year and this one looks almost identical so why don’t we apply the same solution. But from what I have seen here, the problems are, in fact, being handled, each from a fresh viewpoint.