Yes, it's normal practice to generate an ECDSA signature from message $M$ (or it's hash $H(M)$ ), private key $d_U$ and curve parameters, without being given a nonce as input. The nonce $k$ is built as part of the signing process, in one of two ways:
- It's generated a secret integer $k$ uniformly in $[1,n)$ using a true random number generator with secret output. That's the standard definition of ECDSA.
- It's generated a secret integer $k$ in $[1,n)$ using a Pseudo Random Function with key $d_U$, applied to $H(M)$ and optionally other data that needs not be secret (such as a timestamp, or/and a random number). That's what RFC6979 does, prescribing a PRF based on HMAC.
Both methods are as secure: in essence, $k$ is a secret in $[1,n)$ that, to attackers who do not know the private key $d_U$, is unknown and, if it was known, would appear to be random (except for the second option if the same $M$ is re-signed and the optional other data repeats or is absent).
The second method has the advantage of not requiring a true random number generator of cryptographic quality. However it uses private key $d_U$ (and worse mixes it with variable data potentially known to the adversary), therefore the PRF must be protected against side-channel attacks.
Only the second method can insure that signing the same message twice with the same private key generates the same signature, which depending on circumstances is desirable, or not.
