While in L2, the ADSL2 system can instantly re-enter L0 and increase to the maximum data rate as soon the user initiates a file download. The L2 entry/exit mechanisms and resulting data rate adaptations are accomplished without any service interruption or even a single bit error, and as such, are not noticed by the user.

The L3 power modem on the other hand, enables overall power savings at both the ATU-C and the remote ADSL transceiver unit (ATU-R) by entering into sleep mode when the connection is not being used for extended periods of time. L3 is a sleep mode where traffic cannot be communicated over the ADSL connection when the user is not online. When the user returns to go on-line the ADSL transceivers require at least 2 to 3 seconds to re-initialize and enter into steady-state communication mode.

L2 Entry, Exit Procedures
One of the important features of power mode state transitions between L2 and L0 is that they are accomplished without any errors or interruption in service. This is possible because ADSL2 uses a new framing method that involves decoupling the ADSL physical layer medium dependant (PMD) layer from the physical medium specific-transmission convergence (PMS-TC) layer. This feature is also known as S-decoupling because S, which is a parameter that specifies the number of PMD frames in a PMS-TC frame and is fixed to integer only values in previous ADSL standards, is allowed to take on non-integer values in ADSL2. In this way, the PMD layer, which controls the data rate of the connection, can be changed in a seamless manner without affecting the higher layers.

When entering into L2, the PMD layer lowers the data and transmission power in order to save energy. Transition from L0 to L2 is controlled by the ADSL2 transceiver in the central office (ATU-C) [Figure 2].

The ADSL2 transceiver determines the traffic requirements based on the number of ATM cells being transmitted over the ADSL connection. There are several algorithms that can be used to accomplish this. For example, the ATU-C can count the number of idle ATM cells that are transmitted over a period of time (e.g. 10 seconds) and based on a predefined threshold for the number of cells determine whether it is time to enter into L2.

The steps for entry into L2 are as follows:

1. ATU-C determines that the data rate requirements have been significantly reduced and that L2 power saving is desirable.
2. The ATU-C sends a "L2 entry request" message to the ADSL2 transceiver at the remote end (ATU-R). This message contains the minimum and maximum data rate requirements during L2 and the minimum and maximum power cutback values in dB.
3. The ATU-R responds by sending an "L2 entry grant" message to the ATU-R. This message contains L2 data parameters, including the new bits/gains/reorder tables, the L2 power cutback value and the power cutback value to be used for the next L2 exit symbol.
5. The ATU-C enters into L2 by sending the SyncFlag symbol, which is an inverted Synchronization Symbol. On the first discrete multitone transform (DMT) symbol that follows the transmission of the SyncFlag symbol, the system starts transmitting at the new L2 data rate and L2 transmission power level. This is accomplished by using the L2 bits/gains/reorder tables and the L2 power cutback value. During L2 the ATU-C stores the L0 transmission parameters to be used when exiting L2 to re-enter into L0.
6. The ATU-R enters into L2 after detecting the SyncFlag symbol On the first DMT symbol that follows the reception of the SyncFlag symbol, the system starts receiving at the new L2 data rate and L2 transmission power level. This is accomplished by utilizing the L2 bits/gains/reorder tables and the L2 power cutback value. During L2 the ATU-R stores the L0 transmission parameters to be used when exiting L2 to re-enter into L0 and turns on the exit sequence detector.

It should be noted that the storage of the L0 transmission parameters that include the L0 bits, gains, and reordering tables places additional memory requirements on L2 transceivers. In particular, ADSL2 transceivers with 256 carriers must have an additional 768 bytes of memory available to store these tables. Furthermore, the transmitter and receiver must be designed so that the transceivers can switch to the new transmission parameters instantly upon detection of the L2 exit sequence.

Existing L2
When exiting L2 and re-entering into L0, the PMD layer increases the data rates and transmission power levels to the previous L0 levels. Transition from L2 to L0 is controlled by the ATU-C or the ATU-R (Figure 3).

The ATU-R can initiate an exit from L2 into L0 in order to perform receiver signal processing functions (e.g. On-line reconfiguration or bit swapping). The ATU-C transceiver can initiate an exit from L2 into L0 when the data rate requirement is increased beyond the L2 capacity. As in the L2 entry case, algorithms can be used that determine the traffic requirements based on the number of idle ATM cells being transmitted over the ADSL connection. The steps for exit from L2 are as follows:

1. A) The ATU-C determines that the data rate requirements have been significantly increased and re-entry into L0 is required. B) The ATU-R determines that re-entry into L0 requires in order to perform signal processing algorithms and sends an "L0 entry" request message to the ATU-C.
2. The ATU-C enters into L0 by transmitting an L2 exit sequence. On the first DMT symbol that follows the transmission of the L2 exit sequence, the stored L0 bits/gain/reorder tables are used for transmission.
4. The ATU-R detects the L2 exit sequence and enters into L0. On the first DMT symbol that follows the detection of the L2 Exit sequence, the stored L0 bits/gain/reorder tables are used for reception.

Detection of the L2 exit sequence and switching to the stored L0 transmission parameters at the ATU-R provides several new challenges for ADSL transceivers. Since the L2 exit sequence can be sent in place of any regular DMT symbol the ATU-R must detect the L2 exit sequence while receiving and demodulating the regular DMT symbols during steady-state operation. Therefore the L2 exit sequence detector is effectively running in parallel with the normal demodulation functions and upon positive detection of the L2 exit sequence the ATU-R must perform the following steps:

1. Discard the L2 exit sequence DMT symbols and not forward them to the PMS-TC since they do not carry real information.
2. Start using the stored L0 bits/gains/reorder tables on the first DMT symbol that follows the last L2 exit sequence symbol. This means that for certain implementations it may be required to put a second L0 bits/gains/reorder tables in hardware for fast switching because it is unlikely that there enough time to load the current L2 bits/gains/reorder tables with the stored L0 transmission values.

Detection of the L2 exit sequence can be performed using several detection algorithms. One popular algorithm involves multi-tone quadrant detection.¹ Since the L2 exit sequence uses a predefined bit pattern to modulate all multicarrier tones using quadrature phase-shift keying (QPSK), a quadrant detector can be used to effectively and efficiently detect the L2 exit sequence.

The quadrant detector counts the number of tones in the received DMT symbol that have a constellation point that is in the same quadrant as the exit sequence symbol. For example, using N tones the quadrant detector declares a detection of the exit sequence if T tones have a constellation point in the same quadrant as the exit sequence symbol.

Selecting the number of tones (N) and the threshold (T) is a tradeoff between the probability of missed detection versus the probability of a false detection. If N is large and if the value of T is close in value to N the probability of a false detection is low. But under these conditions the probability of a missed detection is high because if N-T tones of the first exit sequence symbol are received in error (i.e. the wrong quadrant) the detector will miss the exit sequence.

On the other hand if N is large and T is much less than N, the probability of a missed detection is low because even if N-T tones of the first exit sequence symbol are received in error (i.e. the wrong quadrant) the detector will still detect the exit sequence. However, under these conditions the probability of a false detections is high. If a real data DMT symbol has T tones with constellation points in the same quadrant as the exit symbol the detector will (falsely) declare a detection.

Understanding L3
The L3 power mode is used when the user in not on-line and no traffic is communicated over the ADSL2 connection. During L3 the ATU may be powered or unpowered and therefore obviously the L3 state provides the maximum power saving. However, unlike the L2 power mode, when in L3 the ADSL2 connection will need to go though an initialization process before the connection can be re-established.

Transition from L0 to L3 is controlled by the either the ATU-C or the ATU-R. The steps for entry into L3 are as follows:

1. The requesting ATU sends an "L3 entry request" message to the ATU on the other end.
2. The ATU on the other end responds by sending an "L3 entry grant" message to the requesting ATU.
3. The ATUs enter the standard orderly shutdown procedure and stop transmitting.

Transition from L3 to L0 is controlled by the either the ATU-R or the ATU-C. The ATU initiates the power mode transition by using either the regular initialization procedure or the short initialization procedure. The regular initialization procedure takes approximately 10 to 15 seconds whereas the short initialization procedure takes approximately 2 to 3 seconds. The short initialization procedure relies on storing and re-utilizing transmission parameters from a previous regular initialization to reduce the required training time.

Wrap Up
As carriers continue to push the deployment of ADSL technology, power consumption will play an increasingly important roll in access box designs. Through the ADSL2 specs, the ITU is addressing this concern head on and providing viable power management techniques for broadband equipment designers.

References
1. ITU-T, RN-034: Detection of the flexible semi-stationary Q-modes signal: Robust detection, non-stationary Q-mode detection, and performance results of a simplified detector, Red Bank, New Jersey, 21 - 25 May 2001

About the Author
Marcos Tzannes is the vice president of strategic technology at Aware, Inc. He holds a BSEE from the University of Central Florida and an MSEE from the University of California at Berkeley. Marcos can be reached at marcos@aware.com.