8.5.3 Time-domain simulation of the clarinet

The starting point for the time-domain simulation approach, as applied to the simple clarinet model, is to assume that just inside the mouthpiece there is at least a short length of uniform tube. We can then use a general expression for acoustic pressure of plane waves in a parallel duct:

$$p(x,t) = p_R(t-x/c) + p_L(t+x/c) \tag{1}$$

where $c$ is the speed of sound, and $p_L$ and $p_R$ denote left-travelling and right-travelling waves, respectively. We derived this expression in section 5.4.2 in the context of waves on an ideal string, but we also showed in section 4.1.1 that plane acoustic waves obey the same one-dimensional wave equation as waves on a string.

Using results from section 4.1.1, we can also write down the corresponding contributions to the acoustic volume flow rate: these are

$$Z^{-1}p_R(t-x/c) \mathrm{~~~and~~~} -Z^{-1}p_L(t+x/c) \tag{2}$$

where $Z$ is a constant called the wave impedance (or characteristic impedance) of the tube. It is equal to $c$ times the density of air divided by the cross-sectional area of the tube.

The reed entrance is at $x=0$, so at that point we have

$$p(t)=p_R(t) + p_L(t) \tag{3}$$

and

$$Zv(t)=p_L(t) – p_R(t) . \tag{4}$$

Now we introduce the linear acoustical behaviour of the clarinet tube. Suppose a pressure pulse in the form of a Dirac delta function is sent out from $x=0$ at time $t=0$. The returning pulse will be delayed, inverted and somewhat rounded, so that it looks something like Fig. 1. Define this “reflection function” to be $r(t)$. In order that there should be no permanent, steady difference of mean pressure between the inside and the outside of the tube, linear acoustics requires this function to satisfy

$$\int_0^\infty{r(t) dt} =-1 . \tag{5}$$

Figure 1. A simple reflection function $r(t)$

Next we make use of a key result shown back in section 2.2.8. We can use $r(t)$ to calculate the reflected pressure wave caused by any outgoing pressure wave by the process of convolution. The result is that

$$p_L(t)=\int_0^\infty{r(\tau) p_R(t-\tau) d \tau} \tag{6}$$

and for brevity we can denote this convolution expression by $p_L=r*p_R$.

Taking the sum and difference of eqs. (3) and (4) gives

$$p+Zv=2 p_L=2r*p_R \tag{7}$$

and

$$p-Zv=2p_R . \tag{8}$$

These can be combined into the form

$$p=p_h + Zv \tag{9}$$

where

$$p_h=r*(p+Zv). \tag{10}$$

The subscript ‘h’ here is to suggest “history”: $p_h$ is something we can easily calculate, by the convolution integral, from a knowledge of the past history of the system. So at a given time step $t$, $p_h$ is a known constant. Equation (9) is then simply the equation of a straight line with slope $1/Z$, crossing the $p$-axis at position $p_h$. But we know that $p$ and $v$ are also connected via the nonlinear function plotted in Fig. 2. So the new values of $p$ and $v$ are found at the intersection between the straight line and the curve: three examples of possible positions of the straight line are indicated in the plot.

Figure 2. The nonlinear characteristic of the mouthpiece (black curve), and three examples of the straight line from eq. (9) with different values of $p_h$.

For the particular curve plotted, and the chosen slope of the straight line, there is always just a single intersection so that the computation is very simple. But if the slope of the line had been lower, so that it was less than the maximum positive slope of the curve, multiple intersections could occur for certain values of $p_h$. We will return to that issue in section ?, in the context of bowed strings, where such multiple intersections have an interesting physical consequence.

The specific parameter values used for the numerical examples of this simple clarinet model in section 8.5 are as follows. They are based on choices made in the original paper describing this method [1]. Units are used in which the characteristic impedance $Z=1$. The smooth portion of the nonlinear curve then has the cubic form

$$v=0.03(p_0 -p)(p-p_0+5)(p-p_0+10) . \tag{11}$$

The maximum positive slope of this curve occurs at the point $p-p_0=-5$, and the value of the slope there is 0.75, so it is indeed lower than $1/Z$.


[1] Michael E. McIntyre, Robert T. Schumacher and Jim Woodhouse; “On the oscillations of musical instruments”, Journal of the Acoustical Society of America 74, 1325–1345 (1983)