The mean spectrum and burst statistics of gravitational waves produced by a cosmological population of cosmic string loops are estimated using analytic approximations, calibrated with earlier simulations. Formulas are derived showing the dependence of observables on the string tension Gµ, in the regime where newly-formed loops are relatively large, not very much smaller than the horizon. Large loops form earlier, are more abundant, and generate a more intense stochastic background and more frequent bursts than assumed in earlier background estimates, enabling experiments to probe lighter cosmic strings of interest to string theory. Predictions are compared with instrument noise from current and future experiments, and with confusion noise from known astrophysical gravitational wave sources such as stellar and massive black hole binaries.In these large-loop models, current data from millisecond pulsar timing already suggests that Gµ is less than about 10 −10 , close to the minimum value where bursts might be detected by Advanced LIGO, and a typical value expected in strings from brane inflation. Because of confusion noise expected from massive black hole binaries, pulsar techniques will not be able to go below about Gµ ≈ 10 −11 . LISA will be sensitive to stochastic backgrounds created by strings as light as Gµ ≈ 10 −15 , at frequencies where it is limited by confusion noise of Galactic stellar populations; however, for those lightest detectable strings, bursts are rarely detectable. For Gµ > 10 −11 , the stochastic background from strings dominates the LISA noise by a large factor, and burst events may also be detectable by LISA, allowing detailed study of loop behavior. Astrophysical confusion might be low enough at 0.1 to 1 Hz to eventually reach Gµ ≈ 10 −20 with future interferometer technology.