- A proposed optical digital-to-analogue converter (oDAC) concept offers several system benefits, including better signal performance, higher bit rates and lower power consumption.
- The oDAC design benefits coherent optics but can also be used in direct-detect designs. This article focusses on coherent optics.
- Coherent system vendors are aware of oDAC technology but it is not part of their current roadmaps.
A 256-QAM constellation using a conventional coherent transmitter (left) and using the oDAC. Note there is no modulator loss (the full area is used) nor any optical warping using the oDAC. Source: Tomkos and Nazarathy.
Systems vendors continue to advance the performance of optical transmission systems. But they are the first to admit the task is getting more complex.
Long-distance transmission is challenging due to the channel impairments introduced by the optical fibre, such as noise, chromatic dispersion, and non-linearities.
Coherent modems have become the established technology that use a powerful digital signal processor (DSP) and optics to counter channel impairments.
In recent years the industry has progressed coherent technology to such a degree that it is now difficult to keep improving optical performance.
One critical component of the coherent DSP is the analogue front end: the transmitter’s digital-to-analogue converters (DACs) and the receiver's analogue-to-digital converters (ADCs).
The DACs take the digital signal input and produce the analogue drive signal for the coherent optics’ Mach-Zehnder modulators. In turm, the DSP’s ADCs sample the signal at the receiver’s optics before recovering the transmitted data payload.
The challenge facing coherent DSP designers is to keep scaling the bandwidth of the DACs and ADCs while maintaining high resolution and high energy efficiency.
This growing challenge has led some researchers to propose alternatives.
One such proposal is an optical digital-to-analogue converter or oDAC.
The status of coherent DSPs
Recent announcements from leading coherent optic vendors, including Cisco's Acacia, Ciena, Infinera, NEL, and Nokia, show the continual progress in hiking the symbol rate of coherent DSPs.
Professor Ioannis Tomkos
Vendors want to keep increasing the symbol rate - the frequency of the symbols where each symbol carries one or more bits, depending on the modulation scheme used - since it remains the best way to reduce the cost of sending network traffic.
First-generation coherent systems used a symbol of 32 gigabaud (GBd). Now, Acacia's currently shipping 5nm CMOS Jannu DSP operates at up to 140GBd. Ciena, meanwhile, has detailed its WaveLogic 6 Extreme, the first coherent DSP implemented in 3nm CMOS that will work at up to 200GBd.
To scale the baud rate, all the sub-systems making up the coherent modem must scale.
The sub-systems include the DSP's DACs and ADCs, the modulator drivers, and the trans-impedance amplifiers. The coherent optics - the coherent driver modulator (CDM) transmitter and the integrated coherent receiver (ICR) - must also scale.
For a 200GBd symbol rate, the bandwidth of all these components must reach 100GHz.
Looking ahead
The industry consensus is that coherent modems will reach 280-300GBd before the decade's end. But to do so will require considerable engineering effort.
The industry offers less visibility after 300+GBd.
System vendors say that at some future point, it will not make economic sense to keep increasing the baud rate. It will be too costly to make the coherent modem and reducing the cost-per-bit will stop.
Already each new generation CMOS node is more costly while new materials are needed to scale the optics. Ciena says it is using silicon photonics for the integrated coherent receiver, while indium phosphide is being used for the transmitter's modulators. Ciena is also looking at thin-film lithium niobate as a modulator technology.
As for DACs and ADCs, circuit designers face considerable challenges in achieving a 100GHz bandwidth.
Moreover, the DACs and ADCs sample faster than the baud rate, typically 1.2x. At OFC, imec, the Belgium technology research centre, outlined its work on 3nm coherent DSPs showing a sample rate of 250 giga-samples/s.
Professor Moshe Nazarathy
Such huge sampling rates explain the interest in optical DACs which can process a high-baud rate signal to generate, using optical parallelism, an ultra-high bit-rate signal based on either multi-level Pulse Amplitude Modulation (PAM) or Quadradure Amplitude Modulation (QAM) signals.
Two prominent professors promoting an optical DAC design are Ioannis Tomkos of the department of electrical and computer engineering at the University of Patras, Greece, and Moshe Nazarathy at the faculty of electrical engineering at the Technion University, Israel.
Limitations of DACs
Tomkos starts by highlighting the shortcomings of conventional DACs.
DACs not only have to operate sampling rates at least as high as the baud rate but they also have a finite resolution. Typically, 6-8 bits are used for coherent designs.
The effective number of bits (ENOB) available are even lower due to the clock jitter when operating the electrical circuits at such high speeds.
The finite effective number of bits limit the use of higher-order modulation schemes. Today, coherent systems use up to 16-ary quadrature amplitude modulation (16-QAM), except for the highest capacity, shortest-distance links.
A second issue is the non-linear nature of the optical modulator’s transfer function. "It’s a sine non-linearity type of response in Mach-Zehnder modulators due to the nature of interference," says Tomkos.
This requires operating the modulator over a reduced range, the linear region of its transfer function around its biasing voltage.
Such curtailing of the driver saves power but results in ‘modulator loss’; the area occupied by the modulator’s constellation points is less than the ideal available (see top left diagram).
"You not driving the modulator to the limit," says Tomkos. "Modulation loss can be as high as 9-12dB which impacts signal recovery at the receiver."
Source: Tomkos and Nazarathy.
The relation between the driving DAC inputs and the discrete optical outputs is generally nonlinear (see diagram above). This means the constellation points look warped and are not spaced equally apart causing signal distortion.
Such optical distortion can be tackled using various specialised DAC architectures but the cost is either higher power, limited speed or extra modulation loss.
"Ideally, we would like to have equal distances between the symbols so we can robustly separate each symbol from the others since we also have electronic errors coming from the DACs that impact the quality of the symbols and shift them from their optimal points," says Tomkos.
The impact of modulation loss and optical distortion also worsen when higher modulation schemes above 16-QAM are used.
The oDAC
Mention the term optical DAC, and specific thoughts come to mind. Is the optical signal sampled? Is the DAC electrical in its input and output, but its inner workings are photonic?
The optical DAC, as proposed by Nazarathy and Tomkos, is neither of the above. Moreover, it uses existing driver electronics based on the simplest traditional lowest-order DACs.
Indeed, the oDAC looks similar to a conventional coherent optics transmitter in terms of components, but the differences in operation and achieved performance are significant.
The oDAC can also be implemented in several ways bringing critical benefits for various system requirements.
Architecture
A conventional coherent optical transmitter splits the incoming laser source and feeds the light equally to the in-phase and quadrature Mach-Zehnder modulators (one arm of which includes a 90-degree phase shifter).
Source: Tomkos and Nazarathy.
The two Mach-Zehnder modulators are driven, as shown. In this example, two drivers implement a bipolar 4-level pulse amplitude modulation (PAM-4) signal such that the coherent transmitter produces a 16-QAM output signal.
The oDAC architecture is subtly different.
The oDAC's main two components are a variable splitter and combiner at the input and output and the Mach-Zehnder modulator pair. Here, both modulators are identical; there is no 90o phase shifter but the differential phase is maintained at 0o degrees and the modulators are operated at full-scale resulting in zero modulation loss (see diagram at the article's start).
Source: Tomkos and Nazarathy.
Each modulator arm is driven by an electrical PAM-4 signal, and the variable splitter-combiner produces the bipolar PAM-16 optical output.
For 16-PAM, 4/5 of the laser signal is fed to one arm and the remaining 1/5 to the other. The PAM-4 DAC drivers for both Mach-Zehnder modulators are identical.
“In the first case, we had 16 symbols in two dimensions (i.e. QAM16); here we have 16 symbols, but in one dimension (i.e. PAM16), the other dimension is missing due to the absence of the 90-degree phase shifter,” says Tomkos.
According to Tomkos, the sine nonlinearity of the optical modulators here is an advantage. "The generated signal does not suffer from modulation loss and optical distortion due to electronic driver mismatch errors, as the noise coming from the electronic DACs gets squelched," he says.
Higher-order modulation
As mentioned, the oDACs can be implemented and arranged in several ways.
For example, two oDACs can be used, one orthogonal in phase to the other, in a conventional coherent transmitter structure to generate a higher modulation signal. For example, two optical DAC arms, each 16-PAM, used as I and Q, will produce a 256-QAM signal.
Source: Tomkos and Nazarathy.
But even more strikingly, more than two parallel modulation paths (by stacking-up more modulators in parallel, see diagram) can be used as an alternative approach to generating higher-order modulation schemes and higher bit rates, and at reduced power consumption per bit.
"The ratio between the bit rate and the baud rate is exactly the number of parallel paths," says Nazarathy. "Another name for it is spectral efficiency: how many bits each symbol carries."
The oDAC uses straightforward drivers. The professors say only PAM-2 or PAM-4 drivers are used. This way, power savings are maximised.
"The big picture is that we offload the electronics burden by going parallel optically," says Nazarathy, adding that what is being traded is electronic DAC complexity and the associated performance limitations of the drivers for optical parallelism of replicated blocks of Mach-Zehnder modulators.
"You don't want to stack things [photonic componentry] serially as if you keep stacking that way, you incur an optical loss because the loss is compounded,” says Nazarathy. Here, the modulators are stacked in parallel, the preferred integration approach.
Moreover, the more paths used, the higher-order the generated optical constellation is. "Eventually, only PAM-2 (Non-return-to-zero) drivers are used and that’s the minimum power consumption you can get," says Tomkos.
"So we have parallelism (at the same laser power) that generates for the same baud rate, double or triple the bit rate [depending on whether 2 or 3 paths are used],” says Nazarathy. And the resulting constellations are near ideal: there is no modulation loss, nor is there optical distortion.
Nazarathy explains such benefits as the result of a ‘divide-and-conquer’ approach.
"If you keep the modulation paths simple, you have more freedom to optimise the drive point of the modulators," he says. "The modulators benefit you more because they are more simply driven." Then, by adding more modulator paths, the system performance improves overall.”
He also notes how the optical implementation is robust to imperfections generated by the electronic circuitry.
____________________________________________________________
Optical DAC: A definition
- Two or more optical modulator units and some static or slowly-tuned ‘glue’ optics.
- The electrical drivers feeding the optical modulators are simple for lowest-power, either PAM2 (NRZ) or PAM4 drivers. Electronic DACs generating higher-order PAM are not needed.
- No high-speed power-hungry digital encoder (mapper) is used. The number of Mach-Zehnder modulators is B where the constellation size is C=2B. This is referred to as Direct Digital Drive. This last condition ensures the lowest power consumption.
____________________________________________________________
Status
The oDAC work is currently at the research stage.
The working of the oDAC has been simulated and verified, and several papers have been published. Patents have also been filed.
At the recent OFC event in San Diago in March, Professor Tomkos met with hyperscalers, systems and components vendors to explain the oDAC technology and its benefits.
The two academics are focused on the oDAC in the optical transmitter, but Nazarathy says they also plan to surprise at the optical receiver end.
Tomkos says the optical DAC is an ideal fit for future coherent transmitters that will be used in 6G networks and datacenter networks, which will carry significant amounts of traffic at ultra-high rates.
The oDAC approach also bodes well for the trend of using linear drive optics. Indeed, the implementation of the oDAC hardware is carried out within the framework of a major R&D project called FLEX-SCALE that Tomkos is co-ordinating and is funded under the first phase of the 6G Smart Networks and Services (SNS) Partnership.
Tomkos believes that the first use of the optical DAC may likely be for data centre interconnect, a more mature market where higher-order modulation formats can be used and low-power is at a premium.
The professors are looking for partners and exploring options to commercialise the technology.