Something that is broadly the same as a coherent 100G QPSK single wavelength optical module, but in two different frequencies, and a passive CWDM mux/demux prism at each end might work. The limitation would be availability of optics for a modern 100G MSA that are both coherent and Tx/Rx at two different THz frequencies.

Or with some box and vendor equipment in between, such as:

http://cdn.extranet.coriant.com/resources/Application-Notes/AN_Groove_Bidirectional_Fiber_74C0169.pdf?mtime=20180206023321

On Tue, Aug 7, 2018 at 1:00 PM Daniel Corbe <dcorbe@hammerfiber.com> wrote:

On 8/7/2018 15:46:03, "Baldur Norddahl" <baldur.norddahl@gmail.com>
wrote:

>Hello
>
>There is a lack of bidirectional one fiber (BIDI) options for 40G and
>100G optics. Usually BIDI is implemented using two CWDM wavelengths,
>one for tx and one for rx. However there is also a lack of CWDM and
>DWDM options for 40G and 100G.
>
>Would it be possible to use an optical circulator like this one
>(customized to 1310 nm)?
>
>https://www.fs.com/de/en/products/33364.html
>
>Combined with a traditional two fiber 1310 nm 10 km 40G QSFP module
>like this: https://www.fs.com/de/en/products/24422.html
>
>The link distance would be 5 km.
>
>The optical circulator separates tx and rx by the direction the light
>travels in. It would work even though both directions use the same
>wavelength. There will likely be some reflection but hopefully
>attenuated enough that it is regarded as background noise.
>
>Has anyone done this? Any reason it would not work?
>
>Regards,
>
>Baldur
>

The main issue you're going to run into (especially trying to plug
anything into a DWDM shelf) is 40G and 100G transceivers usually emit 4
lanes of traffic instead of a single lane like 10 and 1G optics do.

I'd imagine that's why there are so few solutions that don't involve
things like OTN.