Odd numerical issue

[mcg1969]

I've got a simple beamforming problem here:

cvx_begin
        variable w(NA) complex
        minimize(norm(Wstop*w,Inf))
        real(Wzero*w)>=1;
cvx_end

It's proving surprisingly difficult for any solver to handle, though in fact ECOS 1.05 does reasonably well---for awhile. Then it iterates with NaN's:

ECOS 1.0.5 - (c) A. Domahidi, ETH Zurich & embotech 2012-14. Support: ecos@embotech.com

It     pcost         dcost      gap     pres    dres     k/t     mu      step     IR
 0   +0.000e+00   +3.268e+00   +2e+04   6e+01   4e+03   1e+00   2e+00    N/A     1 1 -
 1   +7.680e-01   +1.050e+00   +7e+03   4e+00   3e+02   1e-01   5e-01   0.9158   1 1 1
 2   +3.045e-01   +3.662e-01   +4e+03   7e-01   4e+01   4e-02   2e-01   0.6520   1 1 1
 3   +2.192e-01   +2.469e-01   +2e+03   3e-01   2e+01   2e-02   2e-01   0.4957   1 1 1
 ...
46   +8.559e-02   +8.559e-02   +1e-03   2e-08   1e-06   5e-11   1e-07   0.4098   1 0 1
47   +8.559e-02   +8.559e-02   +9e-04   2e-08   8e-07   4e-11   7e-08   0.7001   1 0 1
48     nan     nan   nan   nan   nan   nan   nan   0.9890   0 0 0
49     nan     nan   nan   nan   nan   nan   nan   0.9890   0 0 0
...
99     nan     nan   nan   nan   nan   nan   nan   0.9890   0 0 0
100     nan     nan   nan   nan   nan   nan   nan   0.9890   0 0 0
Maximum number of iterations reached, recovering best iterate (46) and stopping.

RAN OUT OF ITERATIONS (reached feastol=1.0e-06, reltol=1.7e-02, abstol=1.5e-03).
Runtime: 225.953444 seconds.

A CVX dumpfile showing the inputs and outputs is here.