diff --git a/src/gateware/cxp.py b/src/gateware/cxp.py
index f0cd545..fe8f1e1 100644
--- a/src/gateware/cxp.py
+++ b/src/gateware/cxp.py
@@ -6,6 +6,9 @@ from cxp_downconn import CXP_DownConn_PHY
 from cxp_upconn import CXP_UpConn_PHY
 from cxp_pipeline import *
 
+buffer_depth = 128
+
+@FullMemoryWE()
 class CXP(Module, AutoCSR):
     def __init__(self, refclk, downconn_pads, upconn_pads, sys_clk_freq, debug_sma, pmod_pads):
         self.submodules.upconn = UpConn_Interface(upconn_pads, sys_clk_freq, debug_sma, pmod_pads)
@@ -13,11 +16,76 @@ class CXP(Module, AutoCSR):
         self.submodules.downconn = DownConn_Interface(refclk, downconn_pads, sys_clk_freq, debug_sma, pmod_pads)
         # TODO: support the option high speed upconn
 
+        self.submodules.transmitter = Transmitter()
+
         # TODO: add link layer
 
+    def get_tx_port(self):
+        return self.transmitter.mem.get_port(write_capable=True)
+    
+    def get_mem_size(self):
+        return buffer_depth*downconn_dw
+
+@FullMemoryWE()
+class Transmitter(Module, AutoCSR):
+    def __init__(self):
+        self.cxp_tx_word_len = CSRStorage(bits_for(buffer_depth))
+        self.cxp_tx = CSR()
+
+        # # #
+        
+        self.specials.mem = mem = Memory(downconn_dw, buffer_depth)
+        self.specials.mem_port = mem_port = mem.get_port()
+        self.source = stream.Endpoint(downconn_layout)
+
+
+        tx_done = Signal()
+        addr_next = Signal(bits_for(buffer_depth))
+        addr = Signal.like(addr_next)
+        addr_rst = Signal()
+        addr_inc = Signal()
+
+        # increment addr in the same cycle the moment addr_inc is rise
+        # since memory takes one cycle to shift to the correct addr
+        self.sync += [
+            addr.eq(addr_next),
+            If(self.cxp_tx.re, self.cxp_tx.w.eq(1)),
+            If(tx_done, self.cxp_tx.w.eq(0)),
+        ]
+
+        self.comb += [
+            addr_next.eq(addr),
+            If(addr_rst,
+                addr_next.eq(addr_next.reset),
+            ).Elif(addr_inc,
+                addr_next.eq(addr + 1),
+            ),
+            mem_port.adr.eq(addr_next),
+            self.source.data.eq(mem_port.dat_r)
+        ]
+
+        self.submodules.fsm = fsm = FSM(reset_state="IDLE")
+
+        fsm.act("IDLE",
+            addr_rst.eq(1),
+            If(self.cxp_tx.re, NextState("TRANSMIT"))
+        )
+        fsm.act("TRANSMIT",
+            self.source.stb.eq(1),
+            If(self.source.ack,
+                addr_inc.eq(1),
+            ),
+            If(addr_next == self.cxp_tx_word_len.storage,
+                tx_done.eq(1),
+                NextState("IDLE")
+            )
+        )
+
+        self.submodules.debug_out = debug_out = RX_Debug_Buffer()
+        self.comb += self.source.connect(debug_out.sink)
+
 class DownConn_Interface(Module, AutoCSR):
     def __init__(self, refclk, downconn_pads, sys_clk_freq, debug_sma, pmod_pads):
-
         # # #
 
         self.submodules.phy = phy = CXP_DownConn_PHY(refclk, downconn_pads, sys_clk_freq, debug_sma, pmod_pads)
@@ -111,6 +179,15 @@ class UpConn_Interface(Module, AutoCSR):
         ]
 
 
+        # TODO: rewrite the transmite path into pipeline
+        #
+        # test pak ----+
+        # from gw      |   32                                            32           8
+        #              |---/---> mux -----> trig ack -----> idle word ---/--> conv ---/---> trig -----> PHY
+        #              |                    inserter        inserter                       inserter
+        # data pak ----+
+        # from fw
+
         # Packet FIFOs with transmission priority
         # 0: Trigger packet
         self.submodules.trig = trig = TX_Trigger()