Looking for clarification on the notation described in the MUSIC algorithm in Ralph Schmidt's IEEE paper$^{[1]}$. The data model is:

$$X = AF + W$$

Schmidt defines the following:

X = received signal for each array element
A = steering vector
F = received signal
W = noise

M = number of array elements
D = number linear combination of incident wavefronts

X is dimension M x 1
A is dimension M x D
F is dimension D x 1
W is dimension M x 1


  1. Where is the number of receivers accounted for? A practical system may have a $360^\circ$ element antenna, but only $K$ receivers, where $K \ll M$ (e.g.,$K=8$)
  2. Where is the length of the data incorporated? Assuming $N$ samples for the received signal vector. Intuition leads me to believe that $F$ is of dimension $D\times N$
  3. Does $F$ represent the received signal? If so, what does $X$ represent? Received signal at each element in the array?

$[1]\ $ R. Schmidt, "Multiple emitter location and signal parameter estimation," in IEEE Transactions on Antennas and Propagation, vol. 34, no. 3, pp. 276-280, March 1986


1 Answer 1


Yeah, notation is not ideal.

  1. It is not - he assumes that each of the $M$ antenna elements is connected to its own RF chain, i.e., there are also $M$ receivers available. If you have fewer receivers you need to modify your $A$, it needs to contain the response your $K$ receivers observe given a wave from a certain direction.

  2. He doesn't put it but yeah, $F$ should have $N$ columns (and so should $X$ and $W$) that correspond to $N$ subsequent snapshots (i.e., samples in time) of the received RF data.

  3. The received signal, i.e., the (basedband version of the) RF data at the ouput of the $M$ antennas observed in $N$ subsequent snapshots is the $M \times N$ matrix $X$. The matrix $F$ comprises only the waveforms itself, you could think of it as the signal "before" the antenna array. But that intuition is also slightly misleading. Point is that each antenna is observing a superposition of $D$ wavefronts corresponding to $D$ far-field sources. This can be written as $A F$.

Note that this data model is a gross oversimplification of the actual physics. It ignores near-field sources/reflections, multipath, diffuse components and many things more. It still works well enough in some applications. But the physical interpretation we can put into it has its limits.

  • $\begingroup$ Follow up to 1 - So when $K << M$, should $K$ be used in place of $M$? Follow up to 2 - When you say snapshots for N, do you mean number of receivers, or number of samples in the data? $\endgroup$ Oct 1, 2020 at 18:43
  • 1
    $\begingroup$ I added some clarifications. Does it help? $\endgroup$
    – Florian
    Oct 1, 2020 at 18:56
  • $\begingroup$ Yes, clears up #2. I know that when $K << M$ calibration is required (as you mentioned) to account for the various responses. But still confused how to reconcile $AF$ if $A$ is dimension $M x D$ and $F$ is dimension $D x N$. Neither $A$ nor $F$ is taking into account the number of receivers, $K$. $\endgroup$ Oct 1, 2020 at 20:02
  • $\begingroup$ If you have K receivers your $A$ needs to be $K \times D$... If you still want to use the array steering matrix you need another one in front to reduce $M$ to $K$ in the form of $\Phi A F$. Kinda depends on your architecture, i.e., how are the receivers connected to the antennas? $\endgroup$
    – Florian
    Oct 2, 2020 at 7:12
  • $\begingroup$ Very helpful. Thank you! $\endgroup$ Oct 2, 2020 at 10:37

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.