IQCs for signals with dominant harmonics, i.e., the spectrum is concentrated to be in a certain frequency band. There are three alternatives, 1) bandpass, 2) lowpass, and 3) highpass characteristic:

1. supp \(\hat{f}(j\omega)=[-b,-a]\cup[a,b]\) (supp means support)

2. supp \(\hat{f}(j\omega)=[-a,-a]\), if b=[], a>0

3. supp \(\hat{f}(j\omega)=(-\infty,-|a|]\cup[|a|,\infty)\), if b=[], and a<0





The IQCs are of the form

$$\left\langle f,Mf\right\rangle \geq 0,$$

where \(M=xH^{*}H-d\), and for the three cases we use

1. \(H\) is \(N_{1}\) order Butterworth bandpass filter with cut-off frequencies at \(\omega_{1}=a/N_{2}\) and \(\omega_{2}=bN_{2}\), and

• \( x\textrm{Re}H(ja)-d \geq 0\) 

• \( x\textrm{Re}H(jb)-d \geq 0\) 

• \( x,d \geq 0\)

2. \(H\) is \(N_{1}\) order Butterworth lowpass filter with cut-off frequency at \(\omega=aN_{2}\), and

• \( x\textrm{Re}H(ja)-d \geq 0\) 

• \( x,d \geq 0\)

3. \(H\) is \(N_{1}\) order Butterworth lowpass filter with cut-off frequency at \(\omega=|a|/N_{2}\), and

• \( x\textrm{Re}H(j|a|)-d \geq 0\) 

• \( x,d \leq 0\)

Figure 1 illustrates how “close to optimal” multipliers M look for the three different cases. It is important to note that the choice of parameter \(N_{2}\) can be critical. The reason is that the Butterworth filters are maximaly flat with gain 1/2 at the cut-off frequency. It is therefore useful to use \(N_{2} > 1\). The example in Figure 2 shows that when using \(N_{2} = 1\) the multiplier \(M\) will take large values in the passband while when we use \(N_{2} = 2\) we have a multiplier of the type in Figure 1.

Figure 1: Multipliers for the three cases. An optimal filter has value zero in the passband and minus infinity outside this frequency band.

Figure 2: Let us assume that we are interested in case 2 with a = 1. In solid line we show the multiplier \(M_{1} = H_{1}^{*}H_{1} - d_{1}\), where \(H_{1}\) is a fourth order Butterworth filter with cut-off frequency at \(\omega = 1\) and \(d_{1}\) is chosen such that \(M_{1}(j1) = 0\).  In dashed line we show the filter \(M_{2} = H_{2}^{*}H_{2}-d_{2}\), where \(H_{2}\) is a fourth order Butterworth filter with cut-off frequency at \(\omega = 2\) and \(d_{2}\) is chosen such that \(M_{2}(j1) = 0\). For the first multiplier we used \(N_{2} = 1\) while \(N_{2} = 2\) for the second multiplier. 



n    Size of signal. Default n=1.

a    Frequency bound (lower/upper). Default a=1.

b    Frequency bound (upper). Default b=[].

N1  Order of Butterworth filter.

N2  Determines cut-off frequency of filter.


f    The signal with dominant harmonics.

M  The multiplier M (optional).

x   The parameter x (optional).

d   The parameter d (optional).


We want to compute the L2-gain from f to z for the system \(z=Gf\) when


and f has all its energy in the frequency interval [−5, 5]. The following sequency of commands computes the gain to be gain=1.3343, which is close to the theoretical value \(|G(j5)|=1.3333\).

>>s=tf([1 0],1);






It is interesting to note that if we instead use N2 = 1, i.e., f=iqc_domharmonic(1,5,[],9,1) then we get the gain estimate gain=70.7897. This is expected from the discussion above. In fact, theoretically we obtain the gain \(||G||/\sqrt{2}\) = 70.7116 when N2 = 1.

See also



A. Megretski. Power distribution approach in robust control. In Proceedings of the IFAC Congress, pages 399–402, Sydney, Australia, 1993.