Regenerate initial particles in particle filter

[lwang19-ai]

Hi, I want to ask questions about regenerating particles in particle filter.
I have a generative model running from t1 to t100; the first measurement is at t50.
After I did pf_inialize with the observations at t50, the effective particle is 1. (because of the model dynamics and high dimension)
Hence, I want to regenerate the particles at t0 to get good particles at t50.
I found two ways to do it:

First, I use a selection to select the addresses I want to regenerate or use mh(trace, selection::Selection; ..) to regenerate particles.
However, one question is how to generate a selection with many addresses(more than 100). Do I need to manually add them individually to the Selection object?

Second, I tried to use mh(trace, proposal::GenerativeFunction, proposal_args::Tuple; ..), where I defined a proposal function to regenerate the initial particles. My proposal function is as follows.

@gen function proposal_function_ini(prev_trace, std_a_ini::Float64, std_b_ini::Float64)
    for i in 1:1:Nx
        {(:initial_a => i => :a_ini)} ~ normal(0, std_a_ini)
        {(:initial_b => i => :b_ini)} ~ normal(0, std_b_ini)
    end
end

The addresses like {(:initial_a => I => :a_ini)} are already defined in the generative model used in the particle filter.
But in this case, the algorithm always rejected the regenerated particle, so I got the same trace as the previous trace.
Is there something wrong with this procedure?

Thank you!