1. Introduction

Optical fiber and twisted pair are widely used in fixed broadband access based on the fiber-to-the-cabinet (FTTC) architecture. In the FTTC architecture, twist pair is used for residential access and fiber is utilized for backhauling. To provide higher data rate, optical fiber generally reaches closer to the end users while twisted pair loop length becomes shorter (e.g., <250 m), such as that in the fiber-to-the-drop-point (FTTDp) architecture. G.fast, the newest digital subscriber line (DSL) standard defined by the ITU, is the most promising choice for FTTDp and provides an aggregated throughput of up to 1 Gb/s [1]. However, under the FTTDp architecture, the number of sophisticated remote nodes increases dramatically, and this leads to much increased capital expenditure (Capex) and operational expenditure (Opex). Recently, cloud radio access network (C-RAN) architecture has attracted much attention for wireless networks, as it simplifies remote radio units and provides higher network performance and energy efficiency [2]. C-RAN is essentially a centralized, cloud computing based RAN architecture that can support various generations of wireless communication systems. Optical passive optical network (PON) has been studied for supporting C-RAN [3–7]. Here, we propose and investigate a novel cloud-DSL network architecture consisting of a high-speed twisted-pair access network and a PON-based fronthaul, which moves the major portion of the functionalities of the remote nodes to a central office (CO). This substantially simplifies the remote nodes, reduces both the Capex and the Opex, and enables centralized control and management to improve overall network performance via schemes such as coordinated interference cancellation, in a similar fashion as C-RAN. To enable simple channel aggregation and de-aggregation, we resort to the recently demonstrated electronic code-division-multiple-access (eCDMA) encoding and decoding techniques [8,9]. Based on a proof-of-concept experiment, we further demonstrate the cloud-DSL architecture by transmitting 48 VDSL2 signals over a 20-km standard single-mode fiber (SSMF) fronthaul and a 100-m twisted pair with eCDMA encoding and decoding [8,9] at a CO and a remote node. Using discrete-multi-tone (DMT) with 1024-QAM modulation for each subcarrier, we achieve an aggregated throughput of 5.76 Gb/s with acceptable error-vector magnitude (EVM) performances.

2. Cloud-DSL architecture

Figure 1(a) shows the widely deployed FTTDp architecture that consists of both a fiber backhaul link and a DSL network. At the CO, optical line terminal (OLT) transmits a high-speed optical signal through a fiber and an optical splitter connecting multiple remotes nodes. When the signal reaches a remote node, a DSL access multiplexer (DSLAM) de-modulates the optical signal and modulates the corresponding user data onto multiple DSL signals for transmission to intended customer-premises equipments (CPEs) via twisted pairs. The largest challenge of this network is in the DSLAM, which is a bulky and complex equipment requiring large electrical power and heavy maintenance work and thus significant Opex and Capex for the operators. Figure 1(b) shows the proposed cloud-DSL architecture with centralized processing at the CO and simplified fiber drop points (FDPs), each of which contains a power-efficient eCDMA encoder (ENC) and de-coder (DEC) for uplink and downlink respectively and an analog front end (AFE) circuit to interface with the fiber and the twisted pair. The DSL modulation and demodulation functions are moved to the CO, which allows for centralized processing with high efficiency and performance. For downlink, each DSL signal is encoded by the CDMA encoder with a pre-assigned CDMA code, then all the CDMA encoded DSL signals are summed together to drive a laser to form a modulated optical signal. When the modulated signal reaches a FDP through the fronthaul fiber and optical splitter, it is converted back to the electronic signal by a photo-detector. Then each DSL signal is recovered by correlating with its assigned CDMA code in a CDMA decoder. The decoded DSL signal is further sent to its intended CPE via a dedicated twisted pair. For uplink, the same principle is applied, and a different optical wavelength is used for the uplink transmission than for the downlink transmission.

 

Fig. 1 (a) Current FTTDp architecture with sophisticated remote nodes each containing a DSL access multiplexer (DSLAM) with multiple modulators (Mod) and de-modulators (De-Mod); (b) Proposed cloud-DSL architecture with centralized processing at the CO and simplified fiber drop points (FDPs) each containing a power-efficient eCDMA ENC and DEC pair.

Download Full Size | PDF

In this new architecture, the sophisticated DSLAM, as illustrated in Fig. 1(a), is replaced by a simple eCDMA encoder/decoder (ENC/DEC) module and an AFE circuit. The eCDMA ENC/DEC module can be realized by a small size electrical chip [8,9], and the optical AFE circuits can be a commercially-available compact bidirectional optical transceiver such as a bi-directional optical sub-assembly. It is thus feasible to compress the FDP to a very compact optoelectronic module which reduces the installation and maintenance cost dramatically. Moreover, backward powering is possible due to the low power consumption of the FDP equipment. In addition to large potential savings in Capex and Opex, cloud processing can be realized since all the complex signal processing is now centralized in the CO, which has the complete network parameters. Network optimization can thus be realized using techniques such as network resource sharing and coordination among multiple DSLs.

3. Experimental setup and results

Figure 2(a) shows the experimental setup of the cloud-DSL demonstration. At the CO site, a DSL modulator generates 48 channels of VDSL2 with the 12a profile [10] and a vectoring function is enabled to cancel crosstalk among the VDSL2 signals [11]. The DSL signal is a 12-MHz-bandwidth DMT signal with 4096-QAM or 1024-QAM subcarrier modulation. An eCDMA encoder generates 64 CDMA codes, and we choose 48 codes to transmit 48 users simultaneously. Each DSL user waveform is multiplied by its assigned CDMA code in the time domain. All the 48 CDMA-encoded DSL signals are summed up, before being 4-time interpolated and converted to an analog RF signal by an arbitrary waveform generator (AWG) with 6.25-GSa/s sampling rate and 10-bit resolution. The resulting CDMA chip rate is ¼ of the sampling rate, i.e., 1.5625 GHz, which is 64 times the DMT symbol rate (24.4 MHz). The data rate of each DMT signal is 120 Mb/s ( = 12MHz × 10b/s/Hz) for the case with 1024-QAM. The total throughput of the 48-user eCDMA system using 1024-QAM is thus 5.76 Gb/s. Figure 2(b) shows a single-user CDMA-encoded DMT waveform (green) and the original DMT waveform (yellow), both measured by a real-time oscilloscope (OSC). The the CDMA code period is shown as the time duration between marks 1 and 2. The code period is 40.96 ns, corresponding to the product of the CDMA code length (64) and the inverse of the chip rate. Figure 2(c) shows a measured eCDMA spectrum in the 48-user case. The spectral bandwidth is limited to ~1.5 GHz (the two spectral notches are due to the unused codes).

 

Fig. 2 (a) Experimental setup of the cloud-DSL demonstration; (b) Exemplary CDMA encoded DSL signal of a user; (c) Measured eCDMA spectrum of the 48-user case; (d) Recovered 4096-QAM constellation at the FDP after a single-user transmission with −15 dBm received optical power; (e) Recovered 1024-QAM constellation at the FDP after a 48-user transmission with −15 dBm received power.

Download Full Size | PDF

A directly-modulated laser (DML) with a center wavelength of ~1550 nm, a 3-dB modulation bandwidth of ~2 GHz, a bias current of 60 mA, and a nominal output optical power of 10 dBm is used to convert the eCDMA signal to an optical signal. The optical signal is transmitted through a SSMF whose length is set to 20 km, which is the typical distance in PON [12,13]. A tunable optical attenuator (OATT) emulates the optical splitter typically used in PON. At the receiver site, an avalanche photodiode (APD) module with a combined 3-dB bandwidth of ~2 GHz is used to directly detect the optical signal. In the proof-of-concept experiment, we realize the eCDMA decoding function by using offline digital signal processing (DSP) after analog-to-digital conversion. In real-world applications, the eCDMA encoding and decoding functions can realized by using either analog signal processing [8,9] or DSP, which is being extensively studied for PON [14–17]. The APD module has a built-in transimpedance amplifier with automatic gain control to ensure that the electrical power into a 6.25-GSa/s 8-bit-resolution analog-to-digital converter (ADC) of a real-time OSC is optimized. After being sampled by the ADC, the digital data are sent for CDMA de-coding by offline DSP. In the offline DSP, synchronization (Syn.) is first performed by a training sequence. Then the received signal goes through a splitter that is connected to 48 eCDMA decoders, in each of which every block of 64 samples is correlated with the corresponding CDMA code to produce one CDMA-decoded sample. A switch is used to choose one of the CDMA decoded user data. To investigate the end-to-end performance, we further transmit the decoded signal over a 100-m twisted pair to reach an emulated CPE. We interpolate the decoded signal by 5 times and load it to another AWG with a digital-to-analog converter (DAC) having a sampling rate of 125 MSa/s to regenerate the 12-MHz-bandwidth DMT signal. It is then transmitted over the 100-m twisted pair and sampled by another real-time OSC at 125 MSa/s. The stored samples are processed offline by a data-recovery process consisting of synchronization, time-domain equalization (TEQ), DMT de-modulation, optional forward-error correction (FEC), and signal-to-noise-ratio (SNR) calculation.

Figure 2(d) shows a recovered 4096-QAM constellation with −15 dBm received optical power at the FDP. The final received SNR after the 100-m twist pair is 45 dB, corresponding an EVM of 0.56%, which is acceptable for 4096-QAM [1]. Figure 2(e) shows a recovered 1024-QAM constellation. The final received SNR after the 100-m twist pair is 33 dB, corresponding an EVM of 2.24%, which is acceptable for 1024-QAM [1].

Figure 3(a) shows the 48-user 1024-QAM performance after the end-to-end hybrid transmission over the 20-km SSMF and 100-m twisted pair. All the SNRs meet the requirement for 1024-QAM (32 dB). Figure 3(b) shows the SNRs of the 1st, 25th and 48th users versus the received optical power. The 32-dB SNR requirement is satisfied at −15 dBm optical received power. Since the transmitter optical power is 10 dBm, the optical power budget is 25 dB, meeting the typical GPON optical power budget requirement [12,13]. Remarkably, the total throughput of our DSL fronthaul system is 5.76 Gb/s (with a modulation bandwidth of only ~1.5 GHz). After removing the typical FEC overhead of 7%, the net throughput is 5.38 Gb/s, which is more than twice the GPON capacity [12,13]. Note that with the use of time and wavelength division multiplexed passive optical network (TWDM-PON), more users can be supported. For example, if 4 wavelength channels are used in a TWDM-PON, then this could-DSL system could provide service to 192 users with an average net data rate of ~112 Mb/s per user. Figure 3(c) shows the mean SNR versus the user amount. As the user amount changes from 48 to 1, the SNR increasing from 33 dB to 45 dB. The cloud-DSL can thus schedule the bandwidth resource to improve user experience by simply allocating proper CDMA codes.

 

Fig. 3 (a) Measured 48-user transmission performance; (b) Measured SNR performance vs. the received optical power; (c) Measured mean SNR vs. the number of users.

Download Full Size | PDF

4. Conclusion

We have proposed and experimentally demonstrated a novel ultra-broadband cloud-DSL access network that is enabled by PON-based fronthauling with eCDMA based channel aggregation and de-aggregation. 48 high-speed VDSL2 signals with an aggregated throughput of 5.76 Gb/s is successfully transmitted over a 20-km SSMF and a 100-m twisted pair. This novel solution may greatly reduce the Capex and Opex of future ultra-broadband access, opening new opportunities for converged fiber/copper networks.